Medicaid Expansion and Severe Maternal Morbidity During Delivery Hospitalizations

Abstract

Objectives:

To examine the association between the Affordable Care Act (ACA) Medicaid expansion and severe maternal morbidity (SMM) during the delivery hospitalization.

Methods:

This was a retrospective repeated cross-sectional study using data from the Healthcare Cost and Utilization Project, Agency for Healthcare Research and Quality. All delivery hospitalizations from 2010 to 2018 were analyzed; of the 26 US states included, 16 expanded Medicaid in January 2014 and 10 did not. The primary outcome was the SMM rate, as defined by the Centers for Disease Control and Prevention criteria, exclusive of blood transfusion. We used a quasi-experimental design to compare changes in the SMM rate from pre-expansion (2010-2013) to post-expansion (2015-2018) in states that expanded Medicaid versus those that did not. Generalized synthetic control models were used to estimate the adjusted average treatment effect on the treated.

Results:

The study sample included 11.98 million delivery hospitalizations, of which 82,903 (69.2 per 10,000) experienced SMM. The ACA Medicaid expansion was associated with a 5.2% (95% confidence interval [CI]: 2.1, 8.2) increase in the proportion of delivery hospitalizations paid for by Medicaid, but not with any statistically significant change in the SMM rate (0.02%; 95% CI: −0.01, 0.04). Results were consistent in the subgroup analyses by maternal race and ethnicity and in sensitivity analyses.

Conclusions:

Expanded access to public health insurance was not associated with a reduction in SMM during delivery hospitalizations. Additional clinical and policy interventions may be required to improve maternal health outcomes.

PRECIS:

Expanded access to public health insurance was not associated with a reduction in severe maternal morbidity during the delivery hospitalization.

INTRODUCTION

Addressing severe maternal morbidity and mortality is a public health priority in the United States (US). Severe maternal morbidity (SMM) refers to unintended adverse outcome of pregnancy, labor and delivery resulting in significant short- or long-term consequences to a mother’s health, and SMM is an important precursor of maternal mortality.^{1, 2} The SMM rate increased from 1.35% in 2008 to 2.06% in 2021 in the US, with risk concentrated among African American and Native American women and among women with public insurance coverage.^3–8 Expanding insurance coverage, and thus removing barriers to healthcare access before, during, and after pregnancy, has been proposed as a policy solution to mitigate high and rising rates of SMM.⁹ Up to 60% of women are uninsured at some point from the prepregnancy period to 4-6 months postpartum.^{10, 11}

The Medicaid expansion under the Patient Protection and Affordable Care Act (ACA) provides a natural experiment to assess the association of expanded public insurance coverage with SMM. The expansion gave states the option to expand Medicaid coverage to working-age adults with incomes up to 138% of the Federal Poverty Level (FPL) and started being implemented in January 2014. During the perinatal period, it has led to significant increases in prepregnancy and postpartum coverage and to reductions in insurance discontinuity.^{12, 13} However, evidence regarding the association between Medicaid expansion and SMM during delivery hospitalizations is mixed.^14–16 To further study the effect of this policy change, we use patient-level data in 26 states, including 16 expansion states, and robust causal inference methods to assess the association of the Medicaid expansion with SMM and disparities in SMM during delivery hospitalizations.

METHODS

This was a retrospective repeated cross-sectional study using data from the Healthcare Cost and Utilization Project (HCUP) State Inpatient Databases (SIDs), Agency for Healthcare Research and Quality.¹⁷ SIDs are a census of inpatient discharge information from non-federal hospitals in HCUP-participating states. For each discharge, the SIDs provide some maternal characteristics (e.g., age), one hospital identifier, and patient diagnoses and procedures performed defined using the International Classification of Diseases (ICD), Ninth or Tenth Revision, Clinical Modification.

Intervention in this study was defined as the state implementation of the ACA Medicaid expansion in January 2014, categorized as pre-expansion (from January 2010 to December 2013) or post-expansion period (from January 2015 to December 2018). The ACA Medicaid expansion allowed states to extend coverage to working-age adults with incomes up to 138% of the FPL ($16,105 for a single person in 2014 in the contiguous US, and $ 21,707 for a couple). All the expansion states included in this study expanded Medicaid on January 1^st, 2014.

The exposure was the state ACA Medicaid expansion status, categorized as expansion or non-expansion state. Of the 31 states that provided inpatient data to HCUP between 2010 and 2018, we excluded: (1) two states that expanded Medicaid in 2015 and provided no data before 2015 (AK and IN), (2) two states that covered working-age adults greater than 138% FPL before the implementation of the 2014 ACA Medicaid expansion (DC and VT, “early expanders”), and (3) one non-expansion state (WI) that had a significant increase in Medicaid eligibility for working-age adults during the study period (Appendix 1, available online at http://links.lww.com/xxx).¹⁸ This resulted in a final sample of delivery hospitalizations from 26 states, accounting for approximately 44% of US births: 16 states that expanded Medicaid in January 2014 (AZ, AR, CO, HI, IA, KY, MD, MA, MN, NJ, NM, NY, OR, RI, WA, and WV) and 10 states that did not expand Medicaid between January 2014 and December 2018 (FL, GA, KS, ME, MS, NE, NC, SC, SD, and UT). Among these 26 states included in this study, HI provided no data for 2017 and 2018, and MS no data for 2012.

The primary outcome was the SMM rate, identified using the 20 criteria of the Centers for Disease Control and Prevention (CDC) defined using ICD-9 (10) codes.¹⁹ Blood transfusion was excluded from the definition of SMM.^{20, 21}

As a secondary outcome, we examined the proportion of Medicaid-paid delivery hospitalizations to assess whether the expansion resulted in increased Medicaid coverage at birth in our study sample. However, prepregnancy, prenatal and postpartum coverage (which were most affected by the expansion) are not available in HCUP data.

We identified delivery hospitalizations in the 26 states from January 2010 to December 2018 using a combination of ICD-9 (10) diagnosis and procedure codes.⁷ We excluded mothers who (1) did not give birth in their state of residence as this could lead to misclassification of exposure to Medicaid expansion (2.6%), (2) were aged less than 19 years as they were eligible for Medicaid as children, independent of the Medicaid expansion (6.3%), or were aged greater than 45-years (0.2%), and (3) who gave birth in 2014 (11.2% with 59.6% in expansion states) as these mothers had incomplete exposure to the Medicaid expansion in the prepregnancy period. (Appendix 2, available online at http://links.lww.com/xxx). Since we excluded data for the year 2014, we hypothesized that time from the ACA Medicaid expansion to observable differences in outcomes was one year.

To describe differences between expansion and non-expansion states and between pre- and post-expansion periods, we identified a set of maternal, obstetric, hospital, and state covariates. For statistical adjustment, we limited to covariates associated with the dependent variable of the model (SMM or being covered by Medicaid).

We abstracted the following maternal characteristics from the SID data: age (20-24, 25-29, 30-34, 35-39, or 40-44 years), race and ethnicity, quartile of ZIP code median household income (first quartile: low income), health insurance (Medicare, Medicaid, private, uninsured (self-pay), no charge, or other), and residence based on the Core Based Statistical Areas classification (large metropolitan, small metropolitan, micropolitan, or not metro or micropolitan). Race and ethnicity were categorized as follows: non-Hispanic White, non-Hispanic Black, Hispanic, non-Hispanic Asian or Pacific Islander, non-Hispanic Native American, additional races and ethnicities, and missing. We acknowledge that race is a social construct and we used it as a proxy for racism and discrimination. The following states did not provide information on maternal race and ethnicity for specific time periods: ME (from 2013 to 2018), MN (from 2010 to 2015), NE (from 2010 to 2018), UT (from 2015 to 2017), and WV (from 2010 to 2013). When conducting subgroup analyses according to maternal race and ethnicity, 718,208 (6.0%) discharges from these states and years were excluded.

To measure patient-level obstetric risks, we used ICD-9 (10) codes to calculate the comorbidity index for obstetric patients (CMIOB) and to identify the following obstetric characteristics (Appendix 3, available online at http://links.lww.com/xxx): obesity, previous cesarean delivery, preterm delivery, and delivery mode (vaginal spontaneous, vaginal assisted, or cesarean).^{22, 23}

The following annual hospital-level characteristics were computed using SID data: number of deliveries, cesarean delivery rate, proportion of parturients without comorbidities (i.e., CMIOB = 0), proportion of racial and ethnic minority parturients, proportion of safety net parturients (including Medicare beneficiaries, Medicaid beneficiaries and uninsured), coding intensity (mean number of ICD-9 (10) diagnosis or procedure codes per discharge), and proportion of admissions for delivery during the weekend.

The following annual state characteristics were abstracted from the Area Health Resource File ²⁴: proportion of racial and ethnic minority residents, proportion of unemployed residents, proportion of residents in poverty, proportion of residents with less than high school diploma, and number of obstetricians and gynecologists per 1000 in-hospital births.

We performed statistical analyses using R version 4.3.1 and specific packages (‘tableone’ for the absolute standardized difference, ‘gsynth’ for the generalized synthetic control method, and ‘fixest’ for linear regression model with clustered variance). Analyses were conducted overall and according to maternal race and ethnicity.

We compared characteristics between the pre- and post-expansion periods and between expansion and non-expansion states using the absolute standardized difference (ASD), with a value greater than 0.10 (10%) to define imbalance.²⁵ We estimated the SMM rate overall, by state Medicaid expansion status, and by maternal race and ethnicity.We visually analyzed the temporal trends in the quarterly SMM rate in expansion and in non-expansion states during the pre- and post-expansion periods using plots.

To assess the association between Medicaid expansion and SMM, we used the generalized synthetic control method (GSCM) and patient-level data aggregated at the state- and quarter-level.^26–28 The GSCM is an extension to multiple expansion states of the synthetic control method proposed by Abadie.^{29, 30}For each expansion state, the GSCM constructs a counterfactual weighted combination of non-expansion states (synthetic control) to minimize differences in SMM rates and SMM trends before the expansion. The pre-expansion fit is primarily assessed using visual inspection of the temporal trends in SMM rate in expansion states and in the synthetic control, which should overlap closely during the pre-expansion period (i.e., the difference between expansion states and the synthetic control should be close to zero).

The average treatment effect among the treated (ATT) (i.e. the effect of the Medicaid expansion on the SMM rate in expansion states) is then estimated using the difference in SMM during the post-expansion period between expansion states and the counterfactual synthetic control. The ATT is estimated as the cumulative effect for all post-expansion years (2015-2018). While the GSCM is similar to a difference-in-differences design, it provides some advantages. First, it is robust to violations of the parallel trend assumption (i.e., if pre-expansion trends in SMM differ between expansion and non-expansion states). Second, it can handle unbalanced panels using imputation techniques. In our study sample, two states did not provide data for each year of the study period (HI in 2017 and 2018, and MS in 2012).

To adjustthe ATT, we limited to covariates associated with the dependent variable of the model (SMM rate or proportion of delivery hospitalizations paid for by Medicaid). Covariates associated with the SMM rate included: 1) proportion of parturients 20-24-year-old; 2) proportion of parturients 35-44-year-old; 3) proportion of parturients in quartile 1 of median household income; 4) proportion of parturients residing in large metropolitan area; 5) mean CMIOB; 6) proportion of obese parturients; 7) mean hospital number of deliveries; and 8) mean state number of obstetricians and gynecologists. Covariates associated with the proportion of delivery hospitalizations paid for by Medicaid included: 1) proportion of parturients 20-24-year-old; 2) proportion of parturients in quartile 1 of median household income; 3) proportion of parturients residing in large metropolitan area; 4) state unemployment rate; and 5) state poverty rate. A complete case analysis was performed with 591,099 (4.9%) delivery hospitalizations excluded.

To assess the robustness of our findings, we conducted the following sensitivity analyses. First, we excluded 5 expansion states that already covered working age adults with incomes greater than 0% FPL before the implementation of the Medicaid expansion in January 2014 (AZ, CO, HI, MN and NY) (Appendix 1, available online at http://links.lww.com/xxx). Second, to focus on the low-income sample affected by the expansion, we limited the analysis to Medicaid-paid deliveries. Third, to allow comparability of our findings with previous research, we also estimated a difference-in-differences model.¹⁴

This study was deemed exempt by the Institutional Review Board of the Columbia University Irving Medical Center and is reported according to the STROBE guidelines.³¹

RESULTS

The study sample included 11,978,928 delivery hospitalizations in 26 states (Appendix 2, available online at http://links.lww.com/xxx). Of these, 53.6% were to women aged 20-29-years, 53.3% to non-Hispanic White women, 41.6% to women covered by Medicaid, 26.5% to women living in a low-income neighborhood, and 56.3% to women with a CMIOB equals to zero. Comparisons of expansion and non-expansion states in the pre- and post-expansion periods are reported in Appendix 4 (available online at http://links.lww.com/xxx).

SMM was recorded in 82,903 delivery hospitalizations, yielding a SMM rate of 69.2 per 10,000 (95% CI: 68.7, 69.7) (Table 1). The SMM rate was higher in expansion than in non-expansion states, and in minoritized racial and ethnic women than in White women.

Table 1.

Severe maternal morbidity rate in the 26 states of the study sample during the pre- and post-expansion periods.

	Pre-expansion period (2010-2013)			Post-expansion period (2015-2018)
	No. of women	No. of women with SMM	SMM rate (per 10,000; 95% CI)	No. of women	No. of women with SMM	SMM rate (per 10,000; 95% CI)
All women	5,936,378	41,443	69.8 (69.1, 70.5)	6,042,550	41,460	68.6 (68.0, 69.3)
State Medicaid expansion status
Non-expansion state	2,349,570	16,014	68.2 (67.1, 69.2)	2,464,602	16,300	66.1 (65.1, 67.1)
Expansion state	3,586,808	25,429	70.9 (70.0, 71.8)	3,577,948	25,160	70.3 (69.5, 71.2)
Maternal race and ethnicity (*)
Additional races and ethnicities	260,967	2085	79.9 (76.5, 83.3)	271,104	2119	78.2 (74.8, 81.5)
API	294,066	1894	64.4 (61.5, 67.3)	323,989	2210	68.2 (65.4, 71.0)
Black	864,468	9131	105.6 (103.5, 107.8)	928,555	9888	106.5 (104.4, 108.6)
Hispanic	890,720	6345	71.2 (69.5, 73.0)	960,096	6480	67.5 (65.9, 69.1)
Missing	211,759	1185	56.0 (52.8, 59.1)	158,567	1157	73.0 (68.8, 77.2)
NA	50,101	366	73.1 (65.6, 80.5)	48,076	508	105.7 (96.5, 114.8)
White	2,975,723	18,326	61.6 (60.7, 62.5)	3,022,529	17,317	57.3 (56.4, 58.1)

API

Abbreviations: API: Asian and Pacific Islander; CI: confidence interval; NA: Native American; No.: number; SMM: severe maternal morbidity.

^(*) The following states do not provide information on maternal race and ethnicity for every year of the study period: ME (no data from 2013 to 2018), MN (no data from 2010 to 2015), NE (no data from 2010 to 2018), UT (no data from 2015 to 2017), and WV (no data from 2010 to 2013). Discharges for these states and years are excluded for the computation of the SMM rate according to maternal race and ethnicity.

Temporal trends in the proportion of Medicaid-paid deliveries are reported in Figure 1 and Appendix 5 (available online at http://links.lww.com/xxx). In expansion states, an increase in this proportion was observed during the pre- and post-expansion periods, while a decrease was observed in non-expansion states (Figure 1A). After equalizing the levels and trends during the pre-expansion period using the GSCM (Figure 1B), an increase in the proportion of Medicaid-paid deliveries was observed during the post-expansion period in expansion states and a decrease in the counterfactual synthetic control group. In the adjusted GSCM model, this corresponded to a 5.2 percentage point increase in Medicaid-paid births (95% CI: 2.1, 08.2) in expansion states (Table 2).

Figure 1.

Temporal tends in the quarterly proportion of Medicaid-paid deliveries. The gray vertical rectangle indicates the beginning of the post-expansion period (January 2015). Quarter 1 corresponds to the first quarter of 2010 and quarter 32 to the fourth quarter of 2018. Data for the year 2014 are excluded (washout). A. Trends in the 16 expansion states (black line) and in the 10 nonexpansion states (blue dashed line). B. Trends in the 16 expansion states (black line) and in the counterfactual synthetic control group (blue dashed line).

Table 2.

Crude and adjusted average treatment effect on the treated using the generalized synthetic control method.

	Proportion of Medicaid beneficiaries				Severe maternal morbidity rate
	Crude ATT (95% CI)	P-value	Adjusted ATT (95% CI) (*)	P-value	Crude ATT (95% CI)	P-value	Adjusted ATT (95% CI)(†)	P-value
Main analysis
All women	0.0493(0.0430, 0.0691)	<0.001	0.0517(0.0210, 0.0824)	<0.001	0.0002(−0.0001, 0.0004)	0.121	0.0002(−0.0006, 0.0011)	0.608
Subgroup analysis by maternal race and ethnicity (‡)
Additional races and ethnicities	−0.0432(−0.0711, 0.0074)	0.112	0.0113(−0.0313, 0.0538)	0.604	−0.0005(−0.0067, 0.0057)	0.881	−0.0052(−0.0195, 0.0091)	0.475
API	0.0420(−0.0008, 0.0403)	0.059	0.0645(0.0380, 0.0909)	<0.001	0.0002(−0.0011, 0.0015)	0.803	−0.0004(−0.0038, 0.0031)	0.838
Black	0.0421(0.0016, 0.0405)	0.034	0.0313(0.0066, 0.0559)	0.013	0.0012(0.0002, 0.0021)	0.017	−0.0001(−0.0039, 0.0037)	0.956
Hispanic	0.0425(−0.0064, 0.0564)	0.119	0.0688(0.0291, 0.0109)	<0.001	−0.0002(−0.0009, 0.0004)	0.522	0.0003(−0.0023, 0.0029)	0.816
Missing	0.0415(−0.0138, 0.0436)	0.310	0.0775(0.0413, 0.01136)	<0.001	0.0015(−0.0067, 0.0098)	0.715	0.0056(−0.0047, 0.0159)	0.286
NA	0.0421(0.0041, 0.0383)	0.015	0.0538(0.0313, 0.0763)	<0.001	0.0024(−0.0001, 0.0048)	0.055	−0.0017(−0.0075, 0.0040)	0.552
White	0.0221(0.0012, 0.0429)	0.038	0.0296(0.0062, 0.0531)	0.013	−0.0002(−0.0004, 0.0001)	0.208	0.0002(−0.0005, 0.0009)	0.645

Abbreviations: API: Asian and Pacific Islander; ATT: average treatment effect of the treated; CI: confidence interval; NA: Native American.

^(*) Adjusted for 1) proportion of parturients 20-24-year-old, 2) proportion of parturients in quartile 1 of median household income (low), 3) proportion of parturients residing in large metropolitan area, 4) state unemployment rate, and 5) state poverty rate.

^(†) Adjusted for 1) proportion of parturients 20-24-year-old, 2) proportion of parturients 35-44-year-old, 3) proportion of parturients in quartile 1 of median household income (low), 4) proportion of parturients residing in large metropolitan area, 5) comorbidity index for obstetric patients, 6) proportion of obese parturients, 7) hospital number of deliveries, and 8) state number of obstetricians and gynecologists (per 1000 in-hospital births).

^(‡) For states reporting the information on race and ethnicity, missing values are treated as a category (Missing information on race and ethnicity). The following states does not provide information on maternal race and ethnicity for every year of the study period: ME (no data from 2013 to 2018), MN (no data from 2010 to 2015), NE (no data from 2010 to 2018), UT (no data from 2015 to 2017), and WV (no data from 2010 to 2013). Discharges for these states and years are excluded for the computation of the ATT according to maternal race and ethnicity’

For the SMM rate, there was no obvious visual difference in temporal trends between expansion and non-expansion states during the pre- and post-expansion periods (Figure 2A) or between the expansion states and the counterfactual synthetic control group (Figure 2B). In the adjusted GSCM model,, not statistically significant change in the SMM rate was observed (0.02%; 95% CI: −0.06, 0.11) (Table 2). ATT estimates were consistently null in the subgroup analysis according to maternal race and ethnicity (Table 2) and in all sensitivity analyses (Appendices 6, 7, and 8, available online at http://links.lww.com/xxx).

Figure 2.

Temporal tends in the quarterly severe maternal morbidity rate (SMM). The gray vertical rectangle indicates the beginning of the post-expansion period (January 2015). Quarter 1 corresponds to the first quarter of 2010 and quarter 32 to the fourth quarter of 2018. Data for the year 2014 are excluded (washout). A. Trends in the 16 expansion states (black line) and in the 10 nonexpansion states (blue dashed line). B. Trends in the 16 expansion states (black line) and in the counterfactual synthetic control group (blue dashed line).

DISCUSSION

In this natural experiment study using a robust causal inference technique, we did not find evidence that the ACA Medicaid expansion was associated with a decrease in SMM during delivery hospitalizations, despite an increase in the proportion of Medicaid-paid deliveries.

Gaps in or lack of insurance coverage during the perinatal period have been associated with lower likelihood of receiving early prenatal care and recommended postpartum care, which may lead to poorer maternal outcomes.³² However, previous research on the association of perinatal health insurance coverage with reduced SMM during the delivery hospitalization provides mixed results.³³ In a study in New York State using patient-level hospital discharge data, Medicaid expansion was associated with a small reduction in SMM among women living in low-income neighborhoods relative to those in higher income neighborhoods.¹⁶ In a larger multi-state study, there was no change in SMM in expansion states relative to non-expansion states¹⁴; however, there was no adjustment for patient-level characteristics. In a recent study, implementation of the Medicaid expansion was followed by a decrease in the SMM rate but the study was limited to expansion states and Medicaid beneficiaries.¹⁵ Unlike previous research on Medicaid expansion and SMM, our study analyzed data for a large number of expansion and non-expansion states with adjustment of patient-level characteristics, which are the strongest determinants of SMM.³⁴ However, even after these improvements in study design, we confirm the findings of the prior multi-state study that found no effect of the ACA Medicaid expansion on SMM at childbirth.¹⁴ Importantly, consistent with previous research, the ACA Medicaid expansion resulted in an increase in the proportion of Medicaid-paid deliveries in expansion states.^{13, 35}

The lack of reduction in the SMM rate following the ACA Medicaid expansion has several possible explanations and potential implications for future research on maternal health. First, SMM as defined by the CDC may not be the appropriate outcome to assess the effects of Medicaid expansion on maternal health because most SMM conditions are not easily preventable. A recent study in Maryland reported that only one third of SMM cases, defined as ICU admission or transfusion of 4 units or more of red blood cells, were preventable, and that change in clinician-level factors had the most potential to prevent SMM.³⁶ Second, SMM is only one indicator of maternal health, and we cannot rule out beneficial effects of Medicaid expansion on other indicators. Indeed, there is a need to more clearly define patient-centered measures of maternal health and well-being that can be identified in existing datasets (e.g., discharge, claims, EMR data) and used to evaluate progress on maternal health and the impact of policy and clinical interventions.³⁷ Third, delivery hospitalization may not be the right period of study. Effects of Medicaid policies may be more measurable during the postpartum period, when women are no longer eligible for pregnancy Medicaid. About 20% of new SMM cases occur during the one-year postpartum period.^38–42 Medicaid expansion has been associated with decreased odds of postpartum readmission and postpartum emergency department visits.^43–45 Given that nearly all states have recently implemented postpartum Medicaid extensions - which extend the period of pregnancy Medicaid coverage from 60 days to one year after birth- future research should focus on measuring the health impacts of these investments.

Our findings should be interpreted in the context of several limitations. The exposure measure in our study is susceptible to misclassification. HCUP does not include information on individual household income and thus we could not isolate low-income people in expansion states, affected by the Medicaid expansion. Our definition of exposure (all women in expansion states instead of low-income women in expansion states) is in line with much previous research on the Medicaid expansion during the perinatal period but may contribute to dilute any effects of Medicaid expansion on SMM among those who were Medicaid eligible.^{35, 43, 46} Nevertheless, our results were also consistent in a sensitivity analysis limited to Medicaid-paid childbirths. Our study period includes the adoption of the ICD-10 classification in October 2015, which has been associated with an abrupt change in the SMM rate.^{3, 47} We limited the analysis to HCUP-participating states that expanded Medicaid in January 2014. While these 16 states are likely to be broadly generalizable to all expansion states, it is possible for there to be differential effects in other states. Also, because not all the states included provide a patient identifier, we could not account for women who had more than one childbirth during the study period.

Our findings indicate that expanded access to public health insurance alone was not associated with reduced SMM during delivery hospitalizations. While further study of the effects of ACA implementation beyond the childbirth hospitalization are needed, additional clinical and policy interventions may be required to improve maternal health outcomes.