Keywords
racial/ethnic disparities - in-hospital maternal mortality - cesarean delivery - administrative
database - outcomes research - maternal length of stay
Racial and ethnic disparities in obstetric care and delivery outcomes have been previously
reported, showing black women experience disproportionately higher rates of mortality
and morbidity associated with pregnancy than all other racial and ethnic groups.[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9] In the United States, black women are two to four times more likely than white women
to die in the peripartum period due to cardiovascular, cerebrovascular, and pulmonary
complications.[1]
[3]
[10]
[11]
[12]
[13]
[14]
[15]
In addition, studies have identified black parturients to be more likely than women
from other racial and ethnic groups to have a cesarean delivery, even when controlling
for comorbidities.[1]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25] Cesarean delivery is more deleterious to maternal health than vaginal deliveries
due to an increased risk of poor outcomes with subsequent pregnancies, longer recovery
times, and longer hospital lengths of stay (LOS).[26]
[27] Black women are also more likely to be readmitted after cesarean and vaginal delivery[28] and have longer postpartum hospital LOS.[7]
The disproportionately poor maternal outcomes experienced by black women have been
attributed to the cumulative effects of structural racism.[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36] Demographically, black women are more likely to live in socioeconomically disadvantaged
areas[37] and are less likely to have private insurance as their primary insurance payer than
white women.[38] These factors result in poorer access to quality of care than other racial and ethnic
groups.[39] Additionally, providers' internalized views of their patients' racial or ethnic
identity potentially result in implicit bias leading to a further layer of substandard
care.[33]
[35]
[40] These forces interact in a lived experience of race and ethnicity that affects obstetric
care and influences maternal health outcomes.[41]
Although previous studies have controlled for income and insurance payer as confounders
of maternal outcomes, none have examined them as effect measure modifiers of race
and ethnicity.[7]
[21] Research utilizing data from large, representative data sets is needed to determine
if these differences in outcomes by race and ethnicity, along with the disadvantage
experienced by black women, are due to the unique experience of race and ethnicity
alone, or if the effects of race and ethnicity are modified by primary insurance type
and/or socioeconomic status.[42]
[43]
Using administrative discharge records of more than 6.8 million singleton and multiple
deliveries in five states in the United States over an 8-year period, we aimed to
expand upon existing literature by quantifying differences in maternal outcomes between
racial and ethnic groups by in-hospital mortality, delivery mode, and hospital LOS.
To accomplish this task, we used data from the State Inpatient Database (SID), Healthcare
Cost and Utilization Project (HCUP), and Agency for Healthcare Quality and Research
(AHRQ); a multistate database that approximates a census of inpatient hospital discharges
in participating states.[44] We hypothesized that racial and ethnic differences exist in outcomes in a population
of peripartum patients, and that black patients experience poorer outcomes as compared
with white patients. In particular, we hypothesized that these differences result
in increased rates of in-hospital mortality, cesarean deliveries, and longer LOS,
despite controlling for potential confounders such as present-on-admission comorbidities
or when assessing effect modification (through stratified statistical models) by insurance
payer and socioeconomic status. Our efforts update numbers on outcomes in a peripartum
population, as well as assess the homogeneity of the effects of race/ethnicity on
these outcomes by levels of insurance payer and socioeconomic status.
Materials and Methods
We conducted a retrospective analysis of inpatient singleton and multiple deliveries
from patients aged ≥ 18 years (International Classification of Diseases, Ninth Revision,
Clinical Modification [ICD-9-CM]) using data from the SIDs (SID, HCUP, AHRQ) from
the years 2007 to 2011 for California, and years 2007 to 2014 for Florida, New York,
Maryland, and Kentucky.[44] The SID is an administrative database that contains information on comorbidities,
diagnoses, procedures, and complications of procedures (all of which are identified
by ICD-9-CM codes); an indicator to distinguish present-on-admission comorbidities
and diagnoses, hospital characteristics; patient demographics; dates of admission
and discharge; hospital LOS; and disposition at discharge.[44] Records are by hospital admission.
Singleton deliveries were identified by ICD-9-CM codes V27.0 and V27.1, and multiple
births were identified by ICD-9-CM codes V27.2–8. Deliveries were categorized into
cesarean (ICD-9-CM 74.0–74.2, 74.4, 74.99), operative vaginal (ICD-9-CM 72.0–72.9,
73.3), or normal vaginal deliveries. Ectopic and molar pregnancies, as well as terminated
pregnancies, were excluded from analysis (ICD-9-CM 630–639, 69.01, 75.0).
The Institutional Review Board at Weill Cornell Medicine determined this research
to be exempt from review due to its retrospective nature and lack of patient identifiers.
Outcome Measures
The primary outcome of our study was maternal inpatient mortality. Secondary outcomes
included cesarean delivery versus a vaginal or operative vaginal delivery (i.e., forceps
or vacuum-assisted) and hospital LOS.
Patient and Hospital Variables
The primary variable of interest was race/ethnicity, which was coded as white, black
(Black or African American), Hispanic, and Other (including Asian, Native Hawaiian
or Other Pacific Islander, Native American, and more than one race/ethnicity). Other
patient demographics examined included parturient age; primary payer status (including
Medicare, Medicaid, private insurance, uninsured [including self-pay and no charge],
and Other [including Worker's Compensation, CHAMPUS, CHAMP-VA, Title V, and other
government programs]); delivery type (vaginal, operative vaginal [including forceps
and vacuum deliveries], cesarean); socioeconomic status (operationalized as the median
household income quartile for the patient's ZIP code); hospital state; hospital delivery
volume quartile; and year of delivery.
Present-on-admission comorbidities included the index developed by Bateman et al,[45] a validated comorbidity score for use in obstetric populations.[46] The Bateman index includes weighted measures that predict maternal end-organ injury
or death, including severe preeclampsia or eclampsia, chronic congestive heart failure,
congenital heart disease, pulmonary hypertension, chronic ischemic heart disease,
sickle cell disease, multiple gestation, cardiac valvular disease, systemic lupus
erythematosus, human immunodeficiency virus, mild or unspecified preeclampsia (without
severe preeclampsia or eclampsia), drug abuse, placenta previa, chronic renal disease,
preexisting hypertension, previous cesarean delivery, gestational hypertension, alcohol
abuse, asthma, preexisting diabetes mellitus, and maternal age older than 35 (categorized
as ages 35–39, 40–44, and 45+.
Statistical Analyses
Descriptive statistics by racial/ethnic group and delivery type were calculated separately.
Variables were compared in unadjusted analyses using the appropriate parametric and
nonparametric tests for categorical and ordinal or continuous variables: Pearson's
chi-square test, Fisher's exact test, one-way analysis of variance, Mann–Whitney U-test, or Kruskal–Wallis test.
Multivariate Models
We fit multivariate logistic regression models to predict the outcomes of in-hospital
mortality and cesarean deliveries. We fit multivariate generalized linear models with
a log link that followed a negative binomial distribution to predict the count outcome
of LOS.
Predictor variables in all multivariate models were chosen a priori; all were significantly
different between racial/ethnic groups at α ≤0.05: race (reference: white), primary payer (reference: private insurance), median
household income quartile (reference: first quartile), patient age as a linear term,
the maternal comorbidity index as defined by Bateman,[45] hospital state (reference: Florida), delivery year (reference: 2007), and hospital
delivery volume (reference: first quartile).
For each multivariate model, only observations with nonmissing covariate data were
included. For each outcome, a model was run containing the aforementioned predictor
variables. Each model contained only these predictor variables; a variable indicating
delivery type (reference: vaginal delivery) was additionally included in models for
the outcomes of inpatient mortality and LOS.
Subsequent stratified multivariate models for each outcome were run by subpopulations
of social determinants of health: primary insurance payer and quartile of median income
for the patient's residential ZIP code. These models contained all of the variables
present in the main models, save for the variable by which they were stratified. Stratified
models were used to assess potential heterogeneity in the effects of race/ethnicity
between the main models and within each category of insurance payer and median income
quartile.
For dichotomous outcomes, the area under the receiver operating characteristic curve
was calculated to assess model discrimination. Assumptions regarding the distribution
of the LOS were assessed graphically.
Data are expressed as n (percent), mean (standard deviation), median (interquartile range [IQR]) in bivariate
analyses; odds ratio (OR, 95% confidence interval, CI) for logistic regression models;
or incidence rate ratio (IRR, 95% CI) for generalized linear models with negative
binomial distributions. All p-values are two-sided, and statistical significance is evaluated at the 0.05 α level. Data analysis was conducted in SAS 9.4 (SAS Institute, Cary, NC) and in Stata
SE 15 (StataCorp, College Station, TX).
Results
Sample Attrition
There were 7,109,892 records in the database with an identified singleton or multiple
birth. After dropping cases with missing values on age, the sample size decreased
to 7,097,656. Following the exclusion criteria of Cheesman et al, we omitted cases
of ectopic and molar pregnancies, abortions, dilation and curettage for termination
of pregnancy, aspiration curettage for termination of pregnancy, hysterectomy to terminate
pregnancy, and intra-amniotic injections for abortions, which reduced the sample size
to 7,097,179.[47] The sample was further reduced to include only patients 18 years or older (6,893,416),
with a nonmissing value on gender (6,875,717), a nonmissing final disposition (6,875,593),
nonmissing data on primary insurance payer (6,874,806), and with no more than one
type of preeclampsia coded on the record (including no preeclampsia). This resulted
in a final sample size of 6,872,588.
Descriptive Statistics
A total of 6,872,588 patients were included in the sample; 2,922,680 (42.5%) were
white, 1,930,270 (28.1%) were Hispanic, 957,513 (13.9%) were black, and 871,624 (12.7%)
were of another race/ethnicity ([Table 1]). In addition, 190,501 (2.8%) of the sample had missing values on race/ethnicity.
Black women had the highest percentage of cesarean deliveries, at 38.3%, followed
by Hispanic women, white women, and women of other races/ethnicities.
Table 1
Descriptive statistics by racial and ethnic group
|
Racial or ethnic group
|
Characteristic
|
White
|
Black
|
Hispanic
|
Other
|
Missing
|
Total
|
p-Value
|
Total
|
2,922,680 (42.5)
|
957,513 (13.9)
|
1,930,270 (28.1)
|
871,624 (12.7)
|
190,501 (2.8)
|
6,872,588 (100.0)
|
|
Age in years
|
|
|
|
|
|
|
< 0.0001
|
Mean (standard deviation)
|
29.11 (5.88)
|
27.34 (6.12)
|
27.71 (5.94)
|
30.26 (5.56)
|
28.33 (5.84)
|
28.59 (5.97)
|
|
Delivery type
|
|
|
|
|
|
|
< 0.0001
|
Vaginal delivery
|
1,771,410 (60.6)
|
559,875 (58.5)
|
1,173,464 (60.8)
|
524,081 (60.1)
|
118,153 (62.0)
|
4,146,983 (60.3)
|
|
Operative delivery
|
132,089 (4.5)
|
30,641 (3.2)
|
76,534 (4.0)
|
53,031 (6.1)
|
10,106 (5.3)
|
302,401 (4.4)
|
|
Cesarean delivery
|
1,019,181 (34.9)
|
366,997 (38.3)
|
680,272 (35.2)
|
294,512 (33.8)
|
62,242 (32.7)
|
2,423,204 (35.3)
|
|
Year
|
|
|
|
|
|
|
< 0.0001
|
2007
|
410,157 (14.0)
|
127,018 (13.3)
|
349,142 (18.1)
|
129,047 (14.8)
|
80,739 (42.4)
|
1,096,103 (15.9)
|
|
2008
|
444,927 (15.2)
|
132,896 (13.9)
|
334,990 (17.4)
|
136,837 (15.7)
|
25,246 (13.3)
|
1,074,896 (15.6)
|
|
2009
|
431,493 (14.8)
|
132,176 (13.8)
|
322,959 (16.7)
|
134,240 (15.4)
|
23,155 (12.2)
|
1,044,023 (15.2)
|
|
2010
|
422,804 (14.5)
|
129,814 (13.6)
|
317,772 (16.5)
|
124,554 (14.3)
|
22,829 (12.0)
|
1,017,773 (14.8)
|
|
2011
|
416,470 (14.2)
|
128,537 (13.4)
|
308,151 (16.0)
|
134,059 (15.4)
|
21,591 (11.3)
|
1,008,808 (14.7)
|
|
2012
|
266,118 (9.1)
|
103,250 (10.8)
|
97,495 (5.1)
|
73,616 (8.4)
|
3,711 (1.9)
|
544,190 (7.9)
|
|
2013
|
262,354 (9.0)
|
101,506 (10.6)
|
101,005 (5.2)
|
68,658 (7.9)
|
5,269 (2.8)
|
538,792 (7.8)
|
|
2014
|
268,357 (9.2)
|
102,316 (10.7)
|
98,756 (5.1)
|
70,613 (8.1)
|
7,961 (4.2)
|
548,003 (8.0)
|
|
State
|
|
|
|
|
|
|
< 0.0001
|
California
|
749,996 (25.7)
|
124,694 (13.0)
|
1,152,781 (59.7)
|
327,362 (37.6)
|
88,633 (46.5)
|
2,443,466 (35.6)
|
|
Florida
|
771,684 (26.4)
|
356,888 (37.3)
|
389,102 (20.2)
|
111,575 (12.8)
|
21,405 (11.2)
|
1,650,654 (24.0)
|
|
Kentucky
|
292,165 (10.0)
|
30,044 (3.1)
|
13,679 (0.7)
|
14,008 (1.6)
|
51,349 (27.0)
|
401,245 (5.8)
|
|
Maryland
|
245,526 (8.4)
|
161,971 (16.9)
|
60,226 (3.1)
|
54,505 (6.3)
|
9,937 (5.2)
|
532,165 (7.7)
|
|
New York
|
863,309 (29.5)
|
283,916 (29.7)
|
314,482 (16.3)
|
364,174 (41.8)
|
19,177 (10.1)
|
1,845,058 (26.8)
|
|
Median household income state quartile for patient ZIP code
|
|
|
|
|
|
|
< 0.0001
|
First quartile
|
587,060 (20.1)
|
420,715 (43.9)
|
682,947 (35.4)
|
181,310 (20.8)
|
45,822 (24.1)
|
1,917,854 (27.9)
|
|
Second quartile
|
709,064 (24.3)
|
214,684 (22.4)
|
535,036 (27.7)
|
178,808 (20.5)
|
46,802 (24.6)
|
1,684,394 (24.5)
|
|
Third quartile
|
769,849 (26.3)
|
168,368 (17.6)
|
417,674 (21.6)
|
205,542 (23.6)
|
48,371 (25.4)
|
1,609,804 (23.4)
|
|
Fourth quartile
|
795,561 (27.2)
|
93,112 (9.7)
|
229,905 (11.9)
|
240,172 (27.6)
|
45,368 (23.8)
|
1,404,118 (20.4)
|
|
Missing
|
61,146 (2.1)
|
60,634 (6.3)
|
64,708 (3.4)
|
65,792 (7.5)
|
4,138 (2.2)
|
256,418 (3.7)
|
|
Payer
|
|
|
|
|
|
|
< 0.0001
|
Medicare
|
15,552 (0.5)
|
8,950 (0.9)
|
4,702 (0.2)
|
1,971 (0.2)
|
947 (0.5)
|
32,122 (0.5)
|
|
Medicaid
|
931,436 (31.9)
|
596,250 (62.3)
|
1,276,399 (66.1)
|
325,625 (37.4)
|
79,932 (42.0)
|
3,209,642 (46.7)
|
|
Private insurance
|
1,845,467 (63.1)
|
299,659 (31.3)
|
559,854 (29.0)
|
474,649 (54.5)
|
94,355 (49.5)
|
3,273,984 (47.6)
|
|
Other
|
77,149 (2.6)
|
14,945 (1.6)
|
21,466 (1.1)
|
14,112 (1.6)
|
9,570 (5.0)
|
137,242 (2.0)
|
|
Self-pay/No charge
|
53,076 (1.8)
|
37,709 (3.9)
|
67,849 (3.5)
|
55,267 (6.3)
|
5,697 (3.0)
|
219,598 (3.2)
|
|
Disposition at discharge
|
|
|
|
|
|
|
< 0.0001
|
Routine
|
2,869,876 (98.2)
|
938,663 (98.0)
|
1,908,193 (98.9)
|
854,707 (98.1)
|
187,884 (98.6)
|
6,759,323 (98.4)
|
|
Transfer to short-term hospital
|
2,344 (0.1)
|
687 (0.1)
|
879 (0.0)
|
489 (0.1)
|
159 (0.1)
|
4,558 (0.1)
|
|
Home health care
|
46,739 (1.6)
|
15,320 (1.6)
|
19,183 (1.0)
|
15,301 (1.8)
|
2,195 (1.2)
|
98,738 (1.4)
|
|
Died
|
148 (0.0)
|
134 (0.0)
|
134 (0.0)
|
59 (0.0)
|
13 (0.0)
|
488 (0.0)
|
|
Transfer to other, against medical advice, or destination unknown
|
3,573 (0.1)
|
2,709 (0.3)
|
1,881 (0.1)
|
1,068 (0.1)
|
250 (0.1)
|
9,481 (0.1)
|
|
Hospital volume
|
|
|
|
|
|
|
< 0.0001
|
First quartile
|
832,507 (28.5)
|
164,167 (17.1)
|
498,229 (25.8)
|
144,982 (16.6)
|
70,441 (37.0)
|
1,710,326 (24.9)
|
|
Second quartile
|
657,688 (22.5)
|
222,636 (23.3)
|
571,696 (29.6)
|
224,414 (25.7)
|
49,574 (26.0)
|
1,726,008 (25.1)
|
|
Third quartile
|
690,346 (23.6)
|
289,395 (30.2)
|
475,672 (24.6)
|
221,621 (25.4)
|
40,636 (21.3)
|
1,717,670 (25.0)
|
|
Fourth quartile
|
742,139 (25.4)
|
281,315 (29.4)
|
384,673 (19.9)
|
280,607 (32.2)
|
29,850 (15.7)
|
1,718,584 (25.0)
|
|
Bateman et al (2013) comorbidities
|
|
|
|
|
|
|
|
Alcohol abuse
|
3,665 (0.1)
|
1,685 (0.2)
|
1,243 (0.1)
|
415 (0.0)
|
111 (0.1)
|
7,119 (0.1)
|
< 0.0001
|
Asthma
|
108,244 (3.7)
|
55,663 (5.8)
|
48,648 (2.5)
|
20,121 (2.3)
|
2,964 (1.6)
|
235,640 (3.4)
|
< 0.0001
|
Cardiac valvular disease
|
16,974 (0.6)
|
2,895 (0.3)
|
3,354 (0.2)
|
2,284 (0.3)
|
347 (0.2)
|
25,854 (0.4)
|
< 0.0001
|
Chronic congestive heart failure
|
33 (0.0)
|
40 (0.0)
|
14 (0.0)
|
14 (0.0)
|
< 11 (< 0.0)
|
< 112 (< 0.0)
|
< 0.0001
|
Chronic ischemic heart disease
|
359 (0.0)
|
150 (0.0)
|
125 (0.0)
|
71 (0.0)
|
13 (0.0)
|
718 (0.0)
|
< 0.0001
|
Chronic renal disease
|
6,680 (0.2)
|
2,622 (0.3)
|
4,184 (0.2)
|
1,669 (0.2)
|
180 (0.1)
|
15,335 (0.2)
|
< 0.0001
|
Congenital heart disease
|
2,477 (0.1)
|
515 (0.1)
|
945 (0.0)
|
478 (0.1)
|
54 (0.0)
|
4,469 (0.1)
|
< 0.0001
|
Drug abuse
|
54,310 (1.9)
|
23,706 (2.5)
|
12,990 (0.7)
|
3,929 (0.5)
|
1,396 (0.7)
|
96,331 (1.4)
|
< 0.0001
|
Gestational hypertension
|
93,051 (3.2)
|
31,657 (3.3)
|
39,398 (2.0)
|
15,650 (1.8)
|
2,566 (1.3)
|
182,322 (2.7)
|
< 0.0001
|
Human immunodeficiency virus
|
1,351 (0.0)
|
6,455 (0.7)
|
1,336 (0.1)
|
737 (0.1)
|
76 (0.0)
|
9,955 (0.1)
|
< 0.0001
|
Mild or unspecified preeclampsia
|
69,605 (2.4)
|
38,218 (4.0)
|
45,143 (2.3)
|
17,150 (2.0)
|
2,556 (1.3)
|
172,672 (2.5)
|
< 0.0001
|
Multiple gestation
|
57,638 (2.0)
|
17,740 (1.9)
|
22,338 (1.2)
|
13,462 (1.5)
|
1,671 (0.9)
|
112,849 (1.6)
|
< 0.0001
|
Placenta previa
|
17,023 (0.6)
|
5,548 (0.6)
|
11,335 (0.6)
|
8,298 (1.0)
|
835 (0.4)
|
43,039 (0.6)
|
< 0.0001
|
Preexisting diabetes mellitus
|
22,092 (0.8)
|
12,441 (1.3)
|
19,237 (1.0)
|
7,127 (0.8)
|
819 (0.4)
|
61,716 (0.9)
|
< 0.0001
|
Preexisting hypertension
|
50,718 (1.7)
|
41,760 (4.4)
|
26,277 (1.4)
|
12,522 (1.4)
|
1,807 (0.9)
|
133,084 (1.9)
|
< 0.0001
|
Previous cesarean delivery
|
469,708 (16.1)
|
174,964 (18.3)
|
363,407 (18.8)
|
134,408 (15.4)
|
20,185 (10.6)
|
1,162,672 (16.9)
|
< 0.0001
|
Pulmonary hypertension
|
439 (0.0)
|
520 (0.1)
|
394 (0.0)
|
157 (0.0)
|
26 (0.0)
|
1,536 (0.0)
|
< 0.0001
|
Severe preeclampsia
|
32,371 (1.1)
|
19,765 (2.1)
|
24,263 (1.3)
|
9,429 (1.1)
|
1,480 (0.8)
|
87,308 (1.3)
|
< 0.0001
|
Sickle cell disease
|
3,413 (0.1)
|
5,292 (0.6)
|
1,257 (0.1)
|
4,243 (0.5)
|
258 (0.1)
|
14,463 (0.2)
|
< 0.0001
|
Systemic lupus erythematosus
|
3,480 (0.1)
|
1,760 (0.2)
|
1,964 (0.1)
|
1,044 (0.1)
|
102 (0.1)
|
8,350 (0.1)
|
< 0.0001
|
Bateman et al (2013) comorbidity index
|
|
|
|
|
|
|
< 0.0001
|
Median (Q1; Q3)
|
0 (0; 1)
|
0 (0; 1)
|
0 (0; 1)
|
0 (0; 1)
|
0 (0; 1)
|
0 (0; 1)
|
|
In-hospital death
|
148 (0.0)
|
134 (0.0)
|
134 (0.0)
|
59 (0.0)
|
13 (0.0)
|
488 (0.0)
|
< 0.0001
|
30-d readmission
|
25,492 (1.1)
|
15,668 (2.1)
|
15,719 (1.3)
|
7,012 (1.0)
|
734 (0.6)
|
64,625 (1.3)
|
< 0.0001
|
Length of stay
|
|
|
|
|
|
|
< 0.0001
|
Median (Q1; Q3)
|
2 (2; 3)
|
3 (2; 3)
|
2 (2; 3)
|
2 (2; 3)
|
2 (2; 3)
|
2 (2; 3)
|
|
Total charges in 2016 dollars
|
|
|
|
|
|
|
< 0.0001
|
Median (Q1; Q3)
|
12,204 (7,808; 19,063)
|
12,243 (7,977; 19,705)
|
15,465 (10,240; 23,129)
|
14,375 (9,137; 22,284)
|
11,896 (7,691; 19,247)
|
13,366 (8,533; 20,783)
|
|
Cardiovascular complications grouped variable
|
589 (0.0)
|
379 (0.0)
|
391 (0.0)
|
180 (0.0)
|
27 (0.0)
|
1,566 (0.0)
|
< 0.0001
|
Pulmonary complications grouped variable
|
3,658 (0.1)
|
3,064 (0.3)
|
3,003 (0.2)
|
1,634 (0.2)
|
195 (0.1)
|
11,554 (0.2)
|
< 0.0001
|
Infectious complications grouped variable
|
3,440 (0.1)
|
2,504 (0.3)
|
2,983 (0.2)
|
1,331 (0.2)
|
144 (0.1)
|
10,402 (0.2)
|
< 0.0001
|
Intraoperative complication grouped variable
|
1,196 (0.0)
|
604 (0.1)
|
890 (0.0)
|
368 (0.0)
|
36 (0.0)
|
3,094 (0.0)
|
< 0.0001
|
Digestive system complication
|
333 (0.0)
|
320 (0.0)
|
208 (0.0)
|
108 (0.0)
|
12 (0.0)
|
981 (0.0)
|
< 0.0001
|
Any complication
|
8,680 (0.3)
|
6,358 (0.7)
|
6,987 (0.4)
|
3,425 (0.4)
|
377 (0.2)
|
25,827 (0.4)
|
< 0.0001
|
Abbreviations: IQR, interquartile range; SD, standard deviation.
Note: Percents may not sum to 100 due to rounding and missing values. Continuous variables
analyzed using analysis of variance; categorical variables analyzed using Pearson's
chi-square test or Fisher's exact test. Data provided as n (%), mean (SD), or median (IQR), as appropriate.
Hispanic women had the highest proportion of vaginal deliveries (excluding forceps
or vacuum deliveries), at 60.8%, followed by white women, women of other races/ethnicities,
and black women. Black women were most likely to live in the lowest quartile of median
income, and women of other races/ethnicities were most likely to live in the top quartile
of median income. Two-thirds of Hispanic women were insured by Medicaid, along with
nearly the same percentage of black women. Approximately one-third of women of other
races/ethnicities and white women were insured by Medicaid. White women had the highest
proportion of all racial/ethnic groups of private insurance.
Each racial/ethnic group had the same median scores on the Bateman comorbidity index,
which ranged in values from 0 to 21 (data not shown).
Table 2
Descriptive statistics by delivery type
|
Delivery type
|
Characteristic
|
Vaginal delivery
|
Operative vaginal
|
Cesarean delivery
|
Total
|
p-Value
|
Total
|
4,146,983 (60.3)
|
302,401 (4.4)
|
2,423,204 (35.3)
|
6,872,588 (100.0)
|
|
Age in years
|
|
|
|
|
< 0.0001
|
Mean (standard deviation)
|
28.02 (5.81)
|
27.96 (5.98)
|
29.66 (6.07)
|
28.59 (5.97)
|
|
Race
|
|
|
|
|
< 0.0001
|
White
|
1,771,410 (42.7)
|
132,089 (43.7)
|
1,019,181 (42.1)
|
2,922,680 (42.5)
|
|
Black
|
559,875 (13.5)
|
30,641 (10.1)
|
366,997 (15.1)
|
957,513 (13.9)
|
|
Hispanic
|
1,173,464 (28.3)
|
76,534 (25.3)
|
680,272 (28.1)
|
1,930,270 (28.1)
|
|
Other
|
524,081 (12.6)
|
53,031 (17.5)
|
294,512 (12.2)
|
871,624 (12.7)
|
|
Missing
|
118,153 (2.8)
|
10,106 (3.3)
|
62,242 (2.6)
|
190,501 (2.8)
|
|
Year
|
|
|
|
|
< 0.0001
|
2007
|
665,124 (16.0)
|
55,559 (18.4)
|
375,420 (15.5)
|
1,096,103 (15.9)
|
|
2008
|
651,081 (15.7)
|
51,370 (17.0)
|
372,445 (15.4)
|
1,074,896 (15.6)
|
|
2009
|
627,856 (15.1)
|
47,569 (15.7)
|
368,598 (15.2)
|
1,044,023 (15.2)
|
|
2010
|
614,343 (14.8)
|
45,596 (15.1)
|
357,834 (14.8)
|
1,017,773 (14.8)
|
|
2011
|
609,179 (14.7)
|
44,496 (14.7)
|
355,133 (14.7)
|
1,008,808 (14.7)
|
|
2012
|
325,432 (7.8)
|
20,088 (6.6)
|
198,670 (8.2)
|
544,190 (7.9)
|
|
2013
|
322,713 (7.8)
|
18,978 (6.3)
|
197,101 (8.1)
|
538,792 (7.8)
|
|
2014
|
331,255 (8.0)
|
18,745 (6.2)
|
198,003 (8.2)
|
548,003 (8.0)
|
|
State
|
|
|
|
|
< 0.0001
|
California
|
1,492,973 (36.0)
|
135,345 (44.8)
|
815,148 (33.6)
|
2,443,466 (35.6)
|
|
Florida
|
949,901 (22.9)
|
56,154 (18.6)
|
644,599 (26.6)
|
1,650,654 (24.0)
|
|
Kentucky
|
232,667 (5.6)
|
20,936 (6.9)
|
147,642 (6.1)
|
401,245 (5.8)
|
|
Maryland
|
323,229 (7.8)
|
24,849 (8.2)
|
184,087 (7.6)
|
532,165 (7.7)
|
|
New York
|
1,148,213 (27.7)
|
65,117 (21.5)
|
631,728 (26.1)
|
1,845,058 (26.8)
|
|
Median household income state quartile for patient ZIP code
|
|
|
|
|
< 0.0001
|
First quartile
|
1,178,675 (28.4)
|
77,658 (25.7)
|
661,521 (27.3)
|
1,917,854 (27.9)
|
|
Second quartile
|
1,018,750 (24.6)
|
76,674 (25.4)
|
588,970 (24.3)
|
1,684,394 (24.5)
|
|
Third quartile
|
966,701 (23.3)
|
73,424 (24.3)
|
569,679 (23.5)
|
1,609,804 (23.4)
|
|
Fourth quartile
|
817,882 (19.7)
|
65,452 (21.6)
|
520,784 (21.5)
|
1,404,118 (20.4)
|
|
Missing
|
164,975 (4.0)
|
9,193 (3.0)
|
82,250 (3.4)
|
256,418 (3.7)
|
|
Payer
|
|
|
|
|
< 0.0001
|
Medicare
|
16,489 (0.4)
|
1,150 (0.4)
|
14,483 (0.6)
|
32,122 (0.5)
|
|
Medicaid
|
2,000,245 (48.2)
|
129,712 (42.9)
|
1,079,685 (44.6)
|
3,209,642 (46.7)
|
|
Private insurance
|
1,903,782 (45.9)
|
155,997 (51.6)
|
1,214,205 (50.1)
|
3,273,984 (47.6)
|
|
Other
|
84,793 (2.0)
|
6,403 (2.1)
|
46,046 (1.9)
|
137,242 (2.0)
|
|
Self-pay/No charge
|
141,674 (3.4)
|
9,139 (3.0)
|
68,785 (2.8)
|
219,598 (3.2)
|
|
Disposition at discharge
|
|
|
|
|
< 0.0001
|
Routine
|
4,083,569 (98.5)
|
297,727 (98.5)
|
2,378,027 (98.1)
|
6,759,323 (98.4)
|
|
Transfer to short-term hospital
|
1,483 (0.0)
|
149 (0.0)
|
2,926 (0.1)
|
4,558 (0.1)
|
|
Home health care
|
56,371 (1.4)
|
4,200 (1.4)
|
38,167 (1.6)
|
98,738 (1.4)
|
|
Died
|
110 (0.0)
|
12 (0.0)
|
366 (0.0)
|
488 (0.0)
|
|
Transfer to other, against medical advice, or destination unknown
|
5,450 (0.1)
|
313 (0.1)
|
3,718 (0.2)
|
9,481 (0.1)
|
|
Hospital volume
|
|
|
|
|
< 0.0001
|
First quartile
|
1,047,978 (25.3)
|
82,577 (27.3)
|
579,771 (23.9)
|
1,710,326 (24.9)
|
|
Second quartile
|
1,034,357 (24.9)
|
81,453 (26.9)
|
610,198 (25.2)
|
1,726,008 (25.1)
|
|
Third quartile
|
1,048,627 (25.3)
|
68,085 (22.5)
|
600,958 (24.8)
|
1,717,670 (25.0)
|
|
Fourth quartile
|
1,016,021 (24.5)
|
70,286 (23.2)
|
632,277 (26.1)
|
1,718,584 (25.0)
|
|
Bateman et al (2013) comorbidities
|
|
|
|
|
|
Alcohol abuse
|
4,384 (0.1)
|
226 (0.1)
|
2,509 (0.1)
|
7,119 (0.1)
|
< 0.0001
|
Asthma
|
132,142 (3.2)
|
7,927 (2.6)
|
95,571 (3.9)
|
235,640 (3.4)
|
< 0.0001
|
Cardiac valvular disease
|
13,627 (0.3)
|
1,120 (0.4)
|
11,107 (0.5)
|
25,854 (0.4)
|
< 0.0001
|
Chronic congestive heart failure
|
24 (0.0)
|
< 11 (< 0.0)
|
72 (0.0)
|
< 107 (< 0.0)
|
< 0.0001
|
Chronic ischemic heart disease
|
260 (0.0)
|
39 (0.0)
|
419 (0.0)
|
718 (0.0)
|
< 0.0001
|
Chronic renal disease
|
7,727 (0.2)
|
552 (0.2)
|
7,056 (0.3)
|
15,335 (0.2)
|
< 0.0001
|
Congenital heart disease
|
2,192 (0.1)
|
302 (0.1)
|
1,975 (0.1)
|
4,469 (0.1)
|
< 0.0001
|
Drug abuse
|
58,905 (1.4)
|
3,352 (1.1)
|
34,074 (1.4)
|
96,331 (1.4)
|
< 0.0001
|
Gestational hypertension
|
97,410 (2.3)
|
8,126 (2.7)
|
76,786 (3.2)
|
182,322 (2.7)
|
< 0.0001
|
Human immunodeficiency virus
|
3,845 (0.1)
|
96 (0.0)
|
6,014 (0.2)
|
9,955 (0.1)
|
< 0.0001
|
Mild or unspecified preeclampsia
|
74,957 (1.8)
|
6,517 (2.2)
|
91,198 (3.8)
|
172,672 (2.5)
|
< 0.0001
|
Multiple gestation
|
21,207 (0.5)
|
3,135 (1.0)
|
88,507 (3.7)
|
112,849 (1.6)
|
< 0.0001
|
Placenta previa
|
5,829 (0.1)
|
399 (0.1)
|
36,811 (1.5)
|
43,039 (0.6)
|
< 0.0001
|
Preexisting diabetes mellitus
|
20,426 (0.5)
|
1,557 (0.5)
|
39,733 (1.6)
|
61,716 (0.9)
|
< 0.0001
|
Preexisting hypertension
|
53,498 (1.3)
|
3,319 (1.1)
|
76,267 (3.1)
|
133,084 (1.9)
|
< 0.0001
|
Previous cesarean delivery
|
89,409 (2.2)
|
8,230 (2.7)
|
1,065,033 (44.0)
|
1,162,672 (16.9)
|
< 0.0001
|
Pulmonary hypertension
|
475 (0.0)
|
100 (0.0)
|
961 (0.0)
|
1,536 (0.0)
|
< 0.0001
|
Severe preeclampsia
|
24,111 (0.6)
|
2,151 (0.7)
|
61,046 (2.5)
|
87,308 (1.3)
|
< 0.0001
|
Sickle cell disease
|
7,748 (0.2)
|
613 (0.2)
|
6,102 (0.3)
|
14,463 (0.2)
|
< 0.0001
|
Systemic lupus erythematosus
|
3,978 (0.1)
|
271 (0.1)
|
4,101 (0.2)
|
8,350 (0.1)
|
< 0.0001
|
Bateman et al (2013) comorbidity index
|
|
|
|
|
< 0.0001
|
Median (Q1; Q3)
|
0 (0; 1)
|
0 (0; 1)
|
1 (0; 2)
|
0 (0; 1)
|
|
In-hospital death
|
110 (0.0)
|
12 (0.0)
|
366 (0.0)
|
488 (0.0)
|
< 0.0001
|
30-d readmission
|
27,486 (0.9)
|
2,308 (1.1)
|
34,831 (1.9)
|
64,625 (1.3)
|
< 0.0001
|
Length of stay
|
|
|
|
|
< 0.0001
|
Median (Q1; Q3)
|
2 (2; 3)
|
2 (2; 3)
|
3 (3; 4)
|
2 (2; 3)
|
|
Total charges in 2016 dollars
|
|
|
|
|
< 0.0001
|
Median (Q1; Q3)
|
10,986 (7,414; 16,100)
|
12,693 (8,308; 18,675)
|
19,788 (12,765; 28,280)
|
13,366 (8,533; 20,783)
|
|
Cardiovascular complications grouped variable
|
451 (0.0)
|
52 (0.0)
|
1,063 (0.0)
|
1,566 (0.0)
|
< 0.0001
|
Pulmonary complications grouped variable
|
1,547 (0.0)
|
204 (0.1)
|
9,803 (0.4)
|
11,554 (0.2)
|
< 0.0001
|
Infectious complications grouped variable
|
3,574 (0.1)
|
327 (0.1)
|
6,501 (0.3)
|
10,402 (0.2)
|
< 0.0001
|
Intraoperative complication grouped variable
|
306 (0.0)
|
28 (0.0)
|
2,760 (0.1)
|
3,094 (0.0)
|
< 0.0001
|
Digestive system complication
|
53 (0.0)
|
< 11 (< 0.0)
|
921 (0.0)
|
< 985 (< 0.0)
|
< 0.0001
|
Any complication
|
5,653 (0.1)
|
568 (0.2)
|
19,606 (0.8)
|
25,827 (0.4)
|
< 0.0001
|
Abbreviations: IQR, interquartile range; SD, standard deviation.
Note: Percents may not sum to 100 due to rounding and missing values. Continuous variables
analyzed using analysis of variance; categorical variables analyzed using Pearson's
chi-square test or Fisher's exact test. Data provided as n (%), mean (SD), or median (IQR), as appropriate.
The largest proportion of cesarean deliveries in a population of black parturients
were among those with Medicare or private insurance (45.2 and 42.6%, respectively,
unpublished data), compared with black women insured by Medicaid (36.4%), with other
types of insurance (37.8%), or uninsured (33.6%). By comparison, the largest proportion
of cesarean deliveries in a population of white patients was with patients insured
by Medicare (45.1%), followed by private insurance (36.4%), other insurance (32.2%),
Medicaid (32.1%), and uninsured (29.6%). In a Hispanic patient population, the highest
proportion of those having a cesarean delivery were also on Medicare (46.5%), followed
by private insurance (38.2%), other insurance (34.6%), Medicaid (34.1%), and uninsured
(32.1%). Finally, in a patient population of those from other races, 42.7% of Medicaid
patients had cesarean delivery, followed by 35.5% of patients with private insurance,
31.8% with Medicaid, 34.1% with other types of insurance, and 30.9% of uninsured.
The rate of cesarean deliveries increased with each successive increase in median
income quartile in a black patient population: 37.3% in the first quartile, 39.2%
in the second quartile, 40.1% in the third quartile, and 41.2% in the fourth quartile.
For white patients, the rate of cesarean deliveries followed approximately the same
pattern: 32.8% in the first quartile, 34.2% in the second, 34.8% in the third, and
37.2% in the fourth. Among Hispanic patients, 35.2% of the first income quartile,
35.0% in the second, 35.2% in the third, and 37.6% in the fourth had cesarean deliveries.
Finally, those in other races/ethnicities had the lowest rates of cesarean deliveries
across the board: 31.6% of the first quartile of median income, 34.4% of the second,
34.7% of the third, and 35.1% of the fourth.
Patients with vaginal or operative deliveries had a median score of 0 (IQR: 0; 1)
on the Bateman comorbidity index, while cesarean deliveries had a 1 (IQR: 0; 2) ([Table 2]).
Less than 1% of the women in the entire sample died in-hospital ([Table 1]). In an overall sample, black women had the longest LOS of all racial/ethnic groups.
Multivariate Results
In-Hospital Mortality
After adjustment for patient-level (race [reference: white], primary payer [reference:
private insurance], median household income quartile [reference: first quartile],
patient age as a linear term, the maternal comorbidity index as defined by Bateman,[45] and delivery type [reference: vaginal]) and hospital-level variables (hospital state
[reference: Florida], delivery year [reference: 2007], and hospital delivery volume
[reference: first quartile]), black patients were 90% more likely to die in-hospital
when compared with white women ([Table 3]). Each incremental increase on the Bateman comorbidity index resulted in a 45% increase
in the likelihood of in-hospital maternal mortality (OR: 1.45, 95% CI: 1.39–1.51,
p < 0.001, data not shown). Stratified subgroup analysis by insurance status showed
that black women with Medicaid insurance and black women with private insurance had
increased odds of in-hospital death when compared with white women with those respective
insurances. In a sample of Medicaid patients only, black women were 61% more likely
than white women to die in-hospital; black women and women of other races/ethnicities
(individually) were more than 100% more likely than white women to die in-hospital
in a sample of patients with private insurance. In a population of patients with other
insurance, Hispanic women were more than four times as likely as white women to die
in-hospital.
Table 3
Risk-adjusted outcomes of in-hospital mortality according to race and ethnicity
Population
|
White (reference)
|
Black
|
Hispanic
|
Other
|
Missing
|
N
|
Full sample
|
1.00
|
1.90[c]
(1.47, 2.45)
|
1.17
(0.91, 1.52)
|
1.33
(0.97, 1.82)
|
1.44
(0.81, 2.56)
|
6,872,588
|
Medicare only
|
1.00
|
0.47
(0.08, 2.62)
|
0.56
(0.06, 5.12)
|
1.00
(1.00, 1.00)
|
1.00
(1.00, 1.00)
|
9,823
|
Medicaid only
|
1.00
|
1.61[b]
(1.15, 2.27)
|
0.91
(0.64, 1.30)
|
1.08
(0.66, 1.76)
|
0.75
(0.27, 2.08)
|
3,209,642
|
Private insurance only
|
1.00
|
2.52[c]
(1.65, 3.84)
|
1.49
(0.95, 2.34)
|
2.03[b]
(1.31, 3.15)
|
3.05[b]
(1.45, 6.43)
|
3,273,984
|
Other insurance only
|
1.00
|
1.07
(0.11, 10.08)
|
4.49[a]
(1.21, 16.69)
|
1.00
(1.00, 1.00)
|
1.92
(0.16, 23.38)
|
54,491
|
Uninsured only
|
1.00
|
1.38
(0.48, 3.95)
|
0.68
(0.23, 2.00)
|
0.51
(0.14, 1.90)
|
1.00
(1.00, 1.00)
|
213,901
|
Median income quartile 1 (lowest) only
|
1.00
|
1.45
(0.97, 2.16)
|
0.93
(0.61, 1.43)
|
0.81
(0.41, 1.59)
|
0.84
(0.26, 2.73)
|
1,917,854
|
Median income quartile 2 only
|
1.00
|
1.84[a]
(1.02, 3.33)
|
1.77[a]
(1.04, 3.01)
|
2.72[b]
(1.48, 4.98)
|
1.09
(0.25, 4.65)
|
1,676,265
|
Median income quartile 3 only
|
1.00
|
3.32[c]
(2.01, 5.49)
|
1.01
(0.56, 1.84)
|
2.12[a]
(1.18, 3.81)
|
1.06
(0.25, 4.50)
|
1,609,804
|
Median income quartile 4 (highest) only
|
1.00
|
0.92
(0.35, 2.46)
|
0.98
(0.48, 1.98)
|
0.51
(0.20, 1.33)
|
3.38[a]
(1.28, 8.91)
|
1,399,986
|
Abbreviation: CI, confidence interval.
Note: All numbers reported are odds ratios (95% CI) unless noted otherwise.
a
p < 0.05.
b
p < 0.01.
c
p < 0.001.
In populations of the lowest and highest income quartiles, respectively, no racial/ethnic
group was significantly more or less likely to die in-hospital as compared with white
patients, though this negative finding could be attributed to the small number of
individuals in the entire sample who died in the inpatient setting.
Delivery Type
After adjustment for patient-level (race [reference: white], primary payer [reference:
private insurance], median household income quartile [reference: first quartile],
patient age as a linear term, the maternal comorbidity index as defined by Bateman[45]) and hospital-level variables (hospital state [reference: Florida], delivery year
[reference: 2007], and hospital delivery volume [reference: first quartile]), 6,872,588
women had nonmissing data on all model covariates and were included in our analysis.
Black and Hispanic women were 12 and 23% more likely, respectively, when compared
with white women to have undergone a cesarean delivery as compared with a vaginal
delivery ([Table 4]). Women of other races/ethnicities were also more likely than white women to have
an increased odds of cesarean delivery. For each incremental increase on the Bateman
comorbidity index, women became 2.20 times more likely to have a cesarean delivery
(OR: 2.20, 95% CI: 2.20–2.20, p < 0.001, data not shown).
Table 4
Risk-adjusted outcomes of having a cesarean delivery (vs. vaginal or operative vaginal
delivery) according to race and ethnicity
Population
|
White (reference)
|
Black
|
Hispanic
|
Other
|
Missing
|
N
|
Full sample
|
1.00
|
1.12[c]
(1.12, 1.13)
|
1.23[c]
(1.22, 1.23)
|
1.08[c]
(1.07, 1.08)
|
1.11[c]
(1.09, 1.12)
|
6,872,588
|
Medicare only
|
1.00
|
0.88[c]
(0.83, 0.93)
|
1.07
(0.99, 1.15)
|
0.94
(0.84, 1.03)
|
1.03
(0.89, 1.18)
|
32,122
|
Medicaid only
|
1.00
|
1.14[c]
(1.13, 1.14)
|
1.23[c]
(1.22, 1.24)
|
1.15[c]
(1.14, 1.16)
|
1.12[c]
(1.10, 1.14)
|
3,209,642
|
Private insurance only
|
1.00
|
1.16[c]
(1.15, 1.17)
|
1.20[c]
(1.19, 1.21)
|
1.05[c]
(1.04, 1.06)
|
1.09[c]
(1.07, 1.10)
|
3,273,984
|
Other insurance only
|
1.00
|
1.15[c]
(1.11, 1.20)
|
1.15[c]
(1.11, 1.19)
|
1.06[b]
(1.02, 1.10)
|
1.16[c]
(1.10, 1.23)
|
137,242
|
Uninsured only
|
1.00
|
1.17[c]
(1.13, 1.21)
|
1.26[c]
(1.23, 1.30)
|
1.24[c]
(1.20, 1.28)
|
1.07[a]
(1.00, 1.14)
|
219,598
|
Median income quartile 1 (lowest) only
|
1.00
|
1.15[c]
(1.14, 1.16)
|
1.28[c]
(1.27, 1.29)
|
1.11[c]
(1.10, 1.13)
|
1.11[c]
(1.09, 1.14)
|
1,917,854
|
Median income quartile 2 only
|
1.00
|
1.15[c]
(1.14, 1.16)
|
1.18[c]
(1.17, 1.19)
|
1.11[c]
(1.09, 1.12)
|
1.11[c]
(1.09, 1.13)
|
1,684,394
|
Median income quartile 3 only
|
1.00
|
1.20[c]
(1.18, 1.21)
|
1.23[c]
(1.22, 1.24)
|
1.10[c]
(1.09, 1.12)
|
1.11[c]
(1.08, 1.13)
|
1,609,804
|
Median income quartile 4 (highest) only
|
1.00
|
1.14[c]
(1.12, 1.16)
|
1.19[c]
(1.18, 1.20)
|
1.04[c]
(1.03, 1.05)
|
1.13[c]
(1.10, 1.15)
|
1,404,118
|
Abbreviation: CI, confidence interval.
Note: All numbers reported are odds ratios (95% CI) unless noted otherwise.
a
p < 0.05.
b
p < 0.01.
c
p < 0.001.
Black women, Hispanic women, and women of other races/ethnicities continued to have
increased odds of cesarean delivery compared with white women when stratified subgroup
analysis were performed by primary payer status for Medicaid insurance, private insurance,
other insurance, and in an uninsured population. However, in a population of Medicare
patients, black women had reduced odds of cesarean delivery compared with white women.
Black women, Hispanic women, and women of other races/ethnicities were also individually
more likely than white women to have increased odds of a cesarean delivery in each
quartile of median income, with Hispanic women having the highest odds of a cesarean
delivery as compared with white women in each quartile of median income.
Length of Stay
Models of LOS were interpreted using the IRR, which is the ratio of the average LOS
for one group versus the average LOS for the reference group. In adjusted models (in
which patient- and hospital-level variables were controlled for—including delivery
type), black women had an average LOS 1.10 times greater than white women ([Table 5]). This increase in average LOS was similarly significant in stratified analyses
for each primary insurance payer type and quartile of median income. For each successively
higher median income bracket as compared with the lowest bracket of median income,
the average LOS for black as compared with white women decreased. For each successive
increase in the Bateman comorbidity index, the average LOS increased by 8% (IRR: 1.08,
95% CI: 1.08–1.08, p < 0.001, data not shown).
Table 5
Risk-adjusted outcomes of length of stay (count outcome) according to race and ethnicity
Population
|
White (reference)
|
Black
|
Hispanic
|
Other
|
Missing
|
N
|
Full sample
|
1.00
|
1.10[c]
(1.09, 1.10)
|
1.01[c]
(1.01, 1.01)
|
1.04[c]
(1.04, 1.05)
|
1.01[c]
(1.01, 1.02)
|
6,872,582
|
Medicare only
|
1.00
|
1.10[c]
(1.07, 1.14)
|
1.03
(0.99, 1.07)
|
1.05
(1.00, 1.11)
|
0.98
(0.91, 1.06)
|
32,122
|
Medicaid only
|
1.00
|
1.10[c]
(1.09, 1.10)
|
1.03[c]
(1.02, 1.03)
|
1.05[c]
(1.04, 1.05)
|
1.02[c]
(1.01, 1.03)
|
3,209,640
|
Private insurance only
|
1.00
|
1.10[c]
(1.09, 1.10)
|
0.98[c]
(0.98, 0.98)
|
1.05[c]
(1.04, 1.05)
|
1.01
(1.00, 1.02)
|
3,273,981
|
Other insurance only
|
1.00
|
1.10[c]
(1.08, 1.12)
|
1.02
(1.00, 1.04)
|
1.03[a]
(1.01, 1.05)
|
1.00
(0.97, 1.03)
|
137,242
|
Uninsured only
|
1.00
|
1.13[c]
(1.11, 1.15)
|
1.06[c]
(1.04, 1.07)
|
1.03[c]
(1.01, 1.04)
|
1.03
(0.99, 1.06)
|
219,597
|
Median income quartile 1 (lowest) only
|
1.00
|
1.11[c]
(1.10, 1.12)
|
1.04[c]
(1.03, 1.04)
|
1.05[c]
(1.04, 1.06)
|
1.03[c]
(1.01, 1.04)
|
1,917,854
|
Median income quartile 2 only
|
1.00
|
1.09[c]
(1.09, 1.10)
|
1.01[b]
(1.00, 1.01)
|
1.05[c]
(1.04, 1.05)
|
1.01[a]
(1.00, 1.03)
|
1,684,389
|
Median income quartile 3 only
|
1.00
|
1.09[c]
(1.08, 1.10)
|
1.01[a]
(1.00, 1.01)
|
1.04[c]
(1.04, 1.05)
|
1.01[a]
(1.00, 1.02)
|
1,609,803
|
Median income quartile 4 (highest) only
|
1.00
|
1.08[c]
(1.07, 1.08)
|
1.00
(0.99, 1.00)
|
1.04[c]
(1.03, 1.04)
|
1.00
(0.99, 1.01)
|
1,404,118
|
Abbreviation: CI, confidence interval.
Note: All numbers reported are incidence rate ratios (95% CI) unless noted otherwise.
a
p < 0.05.
b
p < 0.01.
c
p < 0.001.
Hispanic women and women of other races/ethnicities were expected to have slightly
longer LOS than white women, though these findings were not uniform in stratified
analyses by insurance payer and median income quartile.
Discussion
Using comprehensive discharge data from a geographically diverse population in the
United States over an 8-year period, we identified disparities by race and ethnicity
in maternal mortality, delivery type, and LOS.
For each of our outcomes, race/ethnicity served as an independent predictor when controlling
for patient demographics, patient comorbidities specific to an obstetric population,
and hospital-level variables. In stratified models, we analyzed outcomes in subgroups
of insurance payer and median income quartile to assess the homogeneity of the effect
of race/ethnicity across strata. Our findings suggest that race/ethnicity is a significant
predictor of in-hospital mortality, having a cesarean delivery, and having a longer
expected LOS, even in many cases when testing for effect modification by insurance
payer and median income category. Notably, our findings of the homogeneity of the
effect of race/ethnicity in our primary outcome offer confirmatory evidence that black
women across all insurance categories and quartiles of median income experience increased
odds of having a cesarean delivery as compared with white women.
To our knowledge, assessing effect modification by either primary insurance payer
or median income quartile in the relationship between maternal race/ethnicity and
outcomes has hereto not been performed, allowing for an important contribution to
the literature. Moreover, the large, longitudinal sample in our analysis offers a
generalizable insight into the effects of race/ethnicity on maternal outcomes. Perhaps
most importantly, our multivariate analyses adjust for present-on-admission maternal-specific
comorbidities from the Bateman index; adjustment by this index has not been done previously
in a multistate analysis of data from the SID. Even when these comorbidities are taken
into account, black women still experience poorer outcomes compared with white women.
In-Hospital Maternal Mortality/Length of Stay
Our results confirm the findings of a multitude of studies showing racial and ethnic
differences in rates of maternal mortality, with black women being disproportionately
affected.[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[10]
[11]
[12]
[13]
[14]
[15] In our adjusted models, black women were 90% more likely than white women to die
in-hospital; the same was true in stratified analyses by Medicaid and private insurance
populations, with an especially strong increased likelihood of inpatient mortality
in a population of black patients with private insurance. Black women were also more
likely to have an expected slight increase in LOS as compared with white women, even
when controlling for delivery type in adjusted analyses. This corroborated the findings
of Tanner et al, who concluded that black women were more likely than white women
to have LOS longer than 3 days after a scheduled cesarean delivery.[48]
Stratified analyses by payer type and median income quartile are especially advantageous
in examining the relationship between race/ethnicity and LOS, noting the complexities
of the contributing factors to LOS, ranging from patient, provider, and insurance
payer preferences.[49]
Cesarean Delivery
Consistent with the existing literature, we found differences in racial and ethnic
groups in delivery type. There were small but significant differences in the prevalence
of cesarean deliveries across all racial and ethnic groups, with black women having
the highest percentage (38.3%, [Table 1]). In an adjusted logistic regression model, black women were significantly more
likely than white women to undergo a cesarean delivery when controlling for patient-level
comorbidities and hospital characteristics; these estimates were more exaggerated
in a population of Hispanic women ([Table 4]). For black women, these results were maintained in stratified analyses by categories
of insurance payer and median income quartile. Among Hispanic women, results of a
higher likelihood of cesarean delivery as compared with white women were consistent
across all stratified models, save for in a population of Medicare patients. These
estimates of the effects of race/ethnicity on the likelihood of cesarean delivery
in both models with the full sample and stratified by median income quartile and insurance
payer point to the importance of race/ethnicity as a predictor of delivery type.
Explanations about rates of cesarean delivery across racial and ethnic groups vary:
studies have suggested that physician preference, maternal request, and clinical indications
(such as nonreassuring fetal heart tracings or fetal distress) all play a role in
determining the likelihood of a cesarean delivery.[23]
[50]
[51] Roth and Henley noted the advantage of white women and more highly educated women,
both of whom have better access to care and better communication with health care
providers; these populations are less likely to have cesarean deliveries.[23] An additional explanation for some of the racial/ethnic disparities in rates of
cesarean deliveries is based on medical need: black women are more likely to have
an “unnecessary” primary cesarean delivery—when such a procedure is not medically
indicated, but is preferred for financial, medico-legal, or patient demographic factors,
among others.[22]
The epidemiological evidence of racial and ethnic disparities is manifold.
When adjusting for patient characteristics, Chung et al concluded that black women
were 75% more likely than white women to have a cesarean delivery.[20] In models further adding fixed effects of hospital along with random (clustering)
effects of physicians, black women were 54% more likely than white women to undergo
a cesarean delivery. Braveman et al found that black women were 24% more likely than
white women to undergo cesarean deliveries after adjustment for sociodemographic categories
and medical risk factors, as well as hospital characteristics.[18] Roth and Henley added black women were 33% more likely than white women to have
a primary cesarean delivery, even when adjusting for demographic and pregnancy characteristics,
maternal health risks, and complications of labor and delivery.[23] Finally, in adjusted models, Aron et al found that nonwhites had higher rates of
cesarean delivery (OR: 1.34; 95% CI: 1.14–1.57) than whites.[16] Despite these multiple studies, contrary findings do exist. In a nationally representative
sample, Witt et al discovered that there was no effect of race on the likelihood of
both medically indicated and nonmedically indicated cesarean deliveries; indeed, the
strongest predictor was having a previous cesarean delivery.[17] At the provider level, Metz et al found no racial and ethnic differences in cesarean
deliveries between providers who had low, medium, and high rates of primary cesarean
deliveries.[52] These findings suggest that clinical necessity and provider preference may be most
influential in determining delivery type, though mediated by sociodemographic factors.
Limitations
The SID is an administrative database comprised of data from billing records, which
subjects it to concerns of validity. Data may be inaccurately coded or incomplete.
Specific to variables in this study, research to validate coding practices in administrative
databases has shown that delivery type has high sensitivities and positive predictive
values, though select comorbidities do not.[53]
[54] Importantly, the SID does not contain physician-level data to show provider preference
for delivery type and whether this preference was clinically motivated (e.g., due
to vertex presentation); the individual physician's preference has been shown to be
an important factor in the likelihood of a cesarean delivery, net of hospital factors
or patient comorbidities.[52] Nor does the SID contain data on patient preference, which could also affect the
likelihood of a cesarean delivery.[18]
The SID lacks information on the severity of comorbidities, as well as any data on
medications or other markers of prenatal care. Neither does the SID contain behavioral
or social measures on part of the patient or provider. These measures are potentially
relevant factors in explaining the impact of race/ethnicity on our outcomes.[55]
[56] In addition, body mass index, which has been shown to be a significant predictor
of adverse perinatal outcomes,[57]
[58]
[59]
[60]
[61] is not available in the SID.
The unit of analysis in the SID is the inpatient discharge, allowing for multiple
records per patient in the database with multiple pregnancies in the included states
and years.
The obstetric comorbidity index used in our multivariate analyses was designed to
predict maternal death or end-organ injury,[45] although we have included it in multivariate analyses where death or end-organ injury
is not the outcome. Even though this index has been validated,[46] we cannot be certain that it captures the totality of patient conditions that might
contribute to each of our outcomes.
In addition, 2.8% of our sample (190,501) had missing data on race/ethnicity. Because
the percentage of the total sample with missing data on race/ethnicity was below an
arbitrary cutoff point of 5%, we elected not to conduct a sensitivity analysis, despite
our assumption that the data on race and ethnicity were not missing at random.[62] Finally, we were unable to conduct multilevel analyses with hospital as a clustering
variable due to model nonconvergence, though we did include hospital delivery volume
as a covariate in all of our multivariate models.
Strengths
In spite of these limitations, a notable strength of our study is its comprehensive
sample of hospital discharges (approximating a census) over an 8-year period in five
geographically diverse states in the United States. The inclusion of patient and hospital
characteristics as explanatory variables controlled for the effects of preexisting
conditions and for variation in hospital delivery practice.[63]
[64] We also controlled for year of admission to account for changes in laws that have
affected insurance payer reimbursements and coverage, such as the introduction of
the Patient Protection and Affordable Care Act.[65]
Conclusion
This study shows that racial and ethnic disparities in inpatient maternal mortality,
delivery type, and LOS exist, even within strata of insurance payer and income in
models controlling for confounding patient comorbidities and hospital characteristics.
Specifically, racial and ethnic group differences in delivery type exist, with black
women more likely than white women to die in-hospital regardless of primary insurance
payer or residence associated with a particular income bracket. At all levels of health
care, providers should be mindful of the effects of race and ethnicity on maternal
outcomes, irrespective of primary insurance payer or residence in socioeconomic category.
Research into the causal mechanisms of these disparities is needed with the collection
of behavioral and sociological variables in hospital administrative databases.