Insurance coverage and discontinuity during pregnancy: Frequency and associations documented in the PROMISE cohort

Abstract

Objective

To describe insurance patterns and discontinuity during pregnancy, which may affect the experiences of the pregnant person: their timely access to care, continuity of care, and health outcomes.

Data Sources and Study Setting

Data are from the PROMISE study, which utilizes data from community‐based health care organizations (CHCOs) (e.g., federally qualified health centers that serve patients regardless of insurance status or ability to pay) in the United States from 2005 to 2021.

Study Design

This descriptive study was a cohort utilizing longitudinal electronic health record data.

Data Collection/Extraction Methods

Insurance type at each encounter was recorded in the clinical database and coded as Private, Public, and Uninsured. Pregnant people were categorized into one of several insurance patterns. We analyzed the frequency and timing of insurance changes and care utilization within each group.

Principal Findings

Continuous public insurance was the most common insurance pattern (69.2%), followed by uninsured/public discontinuity (11.8%), with 6.4% experiencing uninsurance throughout the entirety of pregnancy. Insurance discontinuity was experienced by 16.6% of pregnant people; a majority of these reflect people transitioning to public insurance. Those with continuous public insurance had the highest frequency of inadequate prenatal care (19.5%), while those with all three types of insurance during pregnancy had the highest percentage of intensive prenatal care (16.5%). The majority (71.7%–81.2%) of those with a discontinuous pattern experienced a single insurance change.

Conclusions

Insurance discontinuity and uninsurance are common within our population of pregnant people seeking care at CHCOs. Our findings suggest that insurance status should be regarded as a dynamic rather than a static characteristic during pregnancy and should be measured accordingly. Future research is needed to assess the drivers of perinatal insurance discontinuity and if and how these discontinuities may affect health care access, utilization, and birth outcomes.

What is known on this topic

• Insurance discontinuity, when a person experiences changes in health insurance including possible periods of uninsurance, is relatively common in the United States and may affect health outcomes.

• Pregnant people are particularly at risk for insurance discontinuity for many reasons including expanded Medicaid eligibility.

• Health insurance discontinuities during pregnancy have been documented using self‐reported, cross‐sectional data, with estimates ranging from 30.1% to 58%.

What this study adds

• We used provider‐recorded longitudinal data to categorize and describe insurance patterns among a population of pregnant people who sought care at community‐based health care organizations.

• 16.6% of pregnant people experienced insurance discontinuity during pregnancy and features such as demographics, prenatal care utilization, and patterns of insurance differed within each insurance group.

• Longitudinal follow‐up of individuals across pregnancy suggests that insurance status during this time should be regarded as a dynamic, rather than a static, feature.

1. INTRODUCTION

The availability of health insurance and access to high‐quality health care is widely assumed to be vital for the health and wellbeing of populations. ¹ , ² Insurance discontinuity occurs when a person experiences changes in health insurance including possible periods of uninsurance (i.e., insurance gaps). ³ , ⁴ , ⁵ Such insurance discontinuities are relatively common in the fragmented United States (US) health care system, and have been found to affect continuity of care, medication adherence, and self‐reported health status. ⁶ , ⁷ Patients are particularly at risk for health insurance discontinuities during and around pregnancy ⁸ , ⁹ , ¹⁰ , ¹¹ for multiple reasons: in particular, becoming eligible for pregnancy‐related Medicaid coverage and, potentially, anticipation of high out‐of‐pocket expenses associated with inpatient childbirth care. ⁶ , ⁹ , ¹¹ , ¹² , ¹³ , ¹⁴

States must, under federal law, provide Medicaid coverage for pregnancy‐related medical services for pregnant individuals with family incomes under 138% of the federal poverty level (FPL). ⁸ States can then extend the FPL threshold to include families with higher incomes through Medicaid or the Children's Health Insurance Program, with thresholds varying between 138% and 380% FPL depending on state and program. ¹⁵ On the other hand, some state Medicaid policies exclude people based on immigration status, such as unauthorized immigrants, legal permanent residents, and those with legal status for fewer than 5 years. ¹⁶ , ¹⁷ In 2021, Medicaid covered 41.0% of all US births. ¹⁸

The existence of health insurance discontinuities (i.e., “churn”) around childbirth is well‐documented. ⁸ , ⁹ , ¹² , ¹³ , ¹⁹ , ²⁰ Prior work has demonstrated that approximately 30% of birthing people in 2009 experienced changes in health insurance coverage during pregnancy. ⁹ Perinatal churn is more common among people with lower income: 83.8% of those who experienced changes in health insurance coverage during pregnancy had incomes ≤200% FPL. ⁹ The prevalence of perinatal churn is also not constant across racial and ethnic groups, disproportionately affecting Indigenous, Hispanic, and Black non‐Hispanic women. ¹⁷ However, these prior studies of perinatal churn have relied on self‐reported and cross‐sectional survey data (e.g., Pregnancy Risk Surveillance and Monitoring System, PRAMS ⁹ , ¹² , ¹³ , ¹⁹ ; Medical Expenditure Panel Survey‐Household Component, MEPS‐HC ⁸ ) to document what is a longitudinal experience: details on the timing and nuances of churn are lacking. These details are necessary to accurately evaluate if and how churn influences perinatal health outcomes. While one study to our knowledge has used longitudinal data to assess insurance churn, ²⁰ the authors were specifically interested in transitions among continuously insured individuals and did not assess periods of uninsurance.

Further, despite policies (such as Medicaid expansion under the Affordable Care Act) aimed at reducing uninsurance in the United States, especially during pregnancy, ²¹ , ²² uninsurance during pregnancy is not uncommon, with 3.9% of people who gave birth in 2021 self‐paying for childbirth care. ¹⁸ Uninsurance despite aforementioned insurance policies is likely driven, in part, by eligibility restrictions for Medicaid based on immigration status. ¹⁶ , ¹⁷

Pregnancy and childbirth are time periods in which timely health care access is strongly related to health outcomes. For example, inadequate prenatal health care is associated with an increased risk of prematurity, stillbirth, neonatal death, and infant death (although specific mechanisms require elucidation). ²³ Insurance disruptions during pregnancy can be expected to delay care and reduce the number of prenatal care (PNC) visits expectant birthing people make during pregnancy, ⁸ even if there is no gap in coverage. ²⁴ However, absent foundational understanding of when, how frequently, and how insurance churn takes place, these potential causal associations cannot be effectively examined.

Health insurance eligibility and care utilization are complex in practice: people can lose or gain insurance multiple times, have multiple forms of insurance, and may not be in possession of the information necessary to accurately report their insurance trajectory due to the nuances and complexities of health insurance in the US. All of the above point toward the need for more fine‐grained data, including longitudinal data, to document the frequency and the nature of discontinuities among those most at risk for insurance gaps and poor outcomes.

Therefore, in this descriptive study, we sought to characterize patterns of pregnancy health insurance using provider‐recorded longitudinal data among a population of women who received care at community‐based health care organizations (CHCOs), many of which were Federally Qualified Health Centers (FQHCs), in the US. FQHCs serve patients who lack insurance, are under‐insured, and other groups that have been economically/socially marginalized, regardless of their ability to pay. The national patient population of FQHCs includes 68% with incomes below the FPL, 63% who identify as a racial or ethnic minority, and 82% who are uninsured or publicly insured. ²⁵

People of all genders experience pregnancy and birth. To acknowledge the gender diversity in the pregnant population, we use both “women” and “persons/people” throughout.

2. METHODS

2.1. Study population

The Preventing Obesity Through Healthy Maternal Gestational Weight Gain in the Safety Net (PROMISE) study cohort is derived from the Accelerating Data Value Across a National Community Health Center Network (ADVANCE) Clinical Research Network, which is funded by the Patient‐Centered Outcomes Research Institute and led by OCHIN, Inc. ²⁶ , ²⁷ ADVANCE integrates outpatient electronic health record (EHR) data for CHCOs from across the US and contains demographic, utilization, and clinical data. OCHIN clinics span multiple states and CHCOs, many of which are FQHCs. The PROMISE study was approved by the Oregon Health & Science University Institutional Review Board.

The PROMISE study identified pregnancies among OCHIN patients by using an electronic algorithm that leverages clinician‐entered pregnancy episodes and outpatient diagnosis and procedure codes. The PROMISE study included pregnancies that started between 16 April, 2004 and 6 July, 2020 in patients 15 years of age or older at pregnancy start and that met the following criteria: (1) gestational age at delivery between 20 and 42 weeks, (2) had ≥1 plausible adult height measurement recorded at ≥16 years of age, and (3) had plausible body mass index and weight measures: ≥1 baseline weight measure (between 365 days prior to and 97 days after pregnancy start), ≥1 weight measure in the second or third trimester, and ≥1 additional weight measure during pregnancy. The PROMISE study contains data from 77,691 pregnancies among 65,245 individuals. Most PROMISE study patients lived in California (38.9%) or Oregon (25.3%) while pregnant, with Massachusetts, Ohio, Washington, and Texas, among others, represented to a smaller degree (see Table S5 for list of states).

2.2. Defining insurance type and discontinuity

Insurance type at each visit was recorded in the clinical database and coded as Public, Private, or Uninsured. Public insurance includes Medicare and Medicaid coverage. Private insurance includes any privately‐ or employer‐purchased insurance plan that is offered by a private health insurance company instead of a state or federal government. Uninsurance includes visits for which no insurer was billed.

Each individual was categorized into one of several mutually exclusive, exhaustive insurance patterns based on the insurance type(s) recorded at their clinical encounters. The first category of insurance pattern was continuous insurance, with the subcategories being continuous private and continuous public. Insurance discontinuity was defined as the presence of more than one insurance type during pregnancy, with the subcategories being private/public discontinuity, uninsured/public discontinuity, uninsured/private discontinuity, and discontinuity between all three types of insurance (hereafter “all‐3‐types discontinuity,” i.e., periods of public, private, and uninsurance). These pattern names reflect the types of insurance present but not the directionality of discontinuity; for example, in the private/public discontinuity pattern, insurance could have changed from private to public or from public to private. The final category of insurance pattern was uninsurance, defined as uninsurance at every visit during pregnancy. This insurance pattern is distinct from the continuous private and continuous public patterns, as it is, by definition, continual lack of insurance rather than any period of insurance, and likely presents distinct challenges and opportunities for intervention.

In order to increase the probability that changes in insurance status reflected actual discontinuities experienced by the patient and not data errors, we corrected apparent insurance discontinuities that had only a single‐visit inconsistency. A single‐visit inconsistency occurred when the insurance type changed for only one visit; after that one visit, the insurance reverted to the insurance type that existed in the penultimate visit (e.g., from public insurance to uninsurance for one visit only and then back to public insurance). If a pregnant person experienced one single‐visit inconsistency as their only discontinuity, the single‐visit inconsistency was replaced with the insurance type at the visits before and after. These individuals were thus assumed to have continuous insurance and were recategorized as such. If a pregnant person experienced more than one instance of a single‐visit inconsistency or any other more complex insurance pattern, their insurance trajectory was not modified. This process resulted in the redesignation of 25,823 visits (3.0% of all visits). A sensitivity analysis, described below, was conducted to explore this decision.

2.3. Exclusions

Visits billed to noncomprehensive insurance plans—such as workers' compensation, dental insurance, and grant/pilot study coverage—were excluded from this analysis. These insurance plans are, by design, only applicable to specific visits, may co‐exist with other insurance plans, and are generally not applicable to pregnancy/PNC. Additionally, visits that were missing for insurance type were excluded. Excluding these visits resulted in the exclusion of 849 (1.1%) pregnancies from the PROMISE cohort. This criterion did not exclude uninsured visits, which were specifically labeled in the EHR as such. Because we were interested in assessing potential changes in insurance type (including multiple changes), which is recorded at each PNC visit, we imposed a minimum number of visits to be eligible for this study. People with fewer than three visits during their pregnancy were thus excluded in order for us to be able to ascertain insurance patterns during pregnancy. This criterion further excluded 1,862 (2.4%) pregnancies, leaving 855,426 visits among 74,980 pregnancies in 62,885 individuals for analysis.

2.4. Descriptive variables

We assessed how sample characteristics, intensity of PNC utilization, and features of insurance discontinuity differed between insurance trajectory groups.

2.5. Descriptive variables: Demographics and health‐related variables

Race and ethnicity were determined from the clinical record and are used in this study as descriptive variables. The categories include Hispanic, White, Black, Asian American/Pacific Islander (AA/PI), Other (American Indian, Alaska Native, Multiple races, “Other” race/ethnicity, and Unknown). Birth order is defined as birth order within the PROMISE dataset; the first pregnancy in the dataset may not be the person's first pregnancy. Income reported at the visit closest to the pregnancy start date was used to categorize income as a percentage of the FPL (<138%, 138–198%, >198%, Missing). We also assessed maternal age at delivery (≤18, 19–24, 25–34, ≥35 years).

2.6. Descriptive variables: Health care utilization

To demonstrate how health care utilization differed between insurance pattern categories, we assessed several measures of utilization. We assessed the observation of PNC in the first trimester, number of PNC visits, amount of time observed, and the adequacy of PNC. The amount of time observed was defined as the time in weeks between the first and last visits. The adequacy of PNC was categorized according to the Revised‐Graduated Prenatal Care Utilization Index (R‐GINDEX). ²⁸ , ²⁹ , ³⁰ , ³¹ The R‐GINDEX utilizes gestational age at delivery, the trimester that PNC began, and the total number of PNC visits made during pregnancy to categorize pregnancies as having inadequate, intermediate, adequate, or intensive PNC utilization. Pregnancy trimesters were defined as follows: <14 weeks (trimester 1), ≥14 to <28 weeks (trimester 2), and ≥28 weeks (trimester 3).

To provide longitudinal detail on insurance trajectories among those experiencing discontinuity, we calculated the duration of coverage and intensity of utilization with each type of insurance. First, we evaluated the amount of time (weeks) and number of visits with each type of insurance. We did not possess the exact date that one's insurance type changed (i.e., interval‐censored data). The cumulative amount of time with each insurance type was thus calculated with the assumption that a given insurance type continued until there was a change at a clinical encounter. We additionally explored the direction of the discontinuity. We report the percentage of those who experienced a single insurance change versus a more complex insurance trajectory, and the mean number of discontinuities experienced. A person was considered to have had a single insurance change if their insurance type changed once and only once. A person was considered to have had a more complex insurance trajectory if they experienced a discontinuous pattern that was not a single insurance change; for example, two insurance changes or multiple single‐visit inconsistencies.

2.7. Sensitivity analysis

To assess how our process of correcting apparent insurance discontinuities that had only a single‐visit inconsistency affected membership in insurance pattern categories and their resulting features, we completed the above descriptive analyses without making this correction. That is, we constructed all of the above variables a second time without making any assumptions about potential data errors and re‐ran all comparisons and analyses.

Descriptive statistics are reported as means with standard deviations, medians with interquartile ranges, or frequencies with percentages. All statistics are reported by the insurance pattern category. Statistical differences between categorical variables were tested with chi‐squared tests, between proportions were tested with z‐tests, between continuous variables were tested with one‐way analysis of variance, and between medians were tested with the Kruskal–Wallis test, all with a level of significance of 0.05. We display our results using a Sankey diagram, with color representing the insurance pattern. ³² Those with an insurance change in more than one direction or in the all‐3‐types group are included in the “More Complex Pattern” designation. To enable visual inspection of discontinuous insurance trajectories, we additionally created graphs charting each person's insurance trajectory as a horizontal line, with insurance type represented in color. ³³

Analyses were conducted using R versions 4.0.2 and 4.3.1. ³⁴

3. RESULTS

3.1. Demographics and health‐related variables

The final study population included 74,980 pregnancies with the following insurance patterns: 5909 continuous private (7.9%), 51,882 continuous public (69.2%), 2385 private/public discontinuous (3.2%), 8871 uninsured/public discontinuous (11.8%), 597 uninsured/private discontinuous (0.8%), 563 with all‐3‐types discontinuity (0.8%), and 4773 uninsured (6.4%) (Table 1 and Figure 1). A discontinuous insurance pattern was seen in 12,416 (16.6%) pregnancies.

TABLE 1.

Characteristics (N [column %]) of the sample of the PROMISE study 2005–2021, overall and among insurance pattern categories (N = 74,980).

	Overall	Continuous		Discontinuous				Uninsured	p Value b
		Private	Public	Private/public	Uninsured/public	Uninsured/private	All‐3‐types
	N = 74,980	N = 5909; 7.9% a	N = 51,882; 69.2% a	N = 2385; 3.2% a	N = 8871; 11.8% a	N = 597; 0.8% a	N = 563; 0.8% a	N = 4773; 6.4% a
Race/ethnicity									<0.001
Hispanic	42,746 (57.0%)	1141 (19.3%)	31,255 (60.2%)	1173 (49.2%)	6023 (67.9%)	176 (29.5%)	263 (46.7%)	2715 (56.9%)
White	17,044 (22.7%)	3674 (62.2%)	9295 (17.9%)	574 (24.1%)	1414 (15.9%)	296 (49.6%)	154 (27.4%)	1637 (34.3%)
Black	8862 (11.8%)	469 (7.9%)	6804 (13.1%)	340 (14.3%)	931 (10.5%)	68 (11.4%)	104 (18.5%)	146 (3.1%)
AA/PI	3502 (4.7%)	287 (4.9%)	2580 (5.0%)	173 (7.3%)	312 (3.5%)	27 (4.5%)	18 (3.2%)	105 (2.2%)
Other	2826 (3.8%)	338 (5.7%)	1948 (3.8%)	125 (5.2%)	191 (2.2%)	30 (5.0%)	24 (4.3%)	170 (3.6%)
Birth order									<0.001
First pregnancy	57,543 (76.7%)	4725 (80.0%)	39,089 (75.3%)	1837 (77.0%)	6847 (77.2%)	514 (86.1%)	465 (82.6%)	4066 (85.2%)
Subsequent pregnancy	17,437 (23.3%)	1184 (20.0%)	12,793 (24.7%)	548 (23.0%)	2024 (22.8%)	83 (13.9%)	98 (17.4%)	707 (14.8%)
FPL (%)									<0.001
<138%	39,596 (52.8%)	1187 (20.1%)	28,502 (54.9%)	1190 (49.9%)	6018 (67.8%)	185 (31.0%)	307 (54.5%)	2207 (46.2%)
138–198%	5543 (7.4%)	502 (8.5%)	3728 (7.2%)	267 (11.2%)	759 (8.6%)	64 (10.7%)	74 (13.1%)	149 (3.1%)
>198%	3965 (5.3%)	1158 (19.6%)	1807 (3.5%)	200 (8.4%)	546 (6.2%)	108 (18.1%)	46 (8.2%)	100 (2.1%)
Missing	25,876 (34.5%)	3062 (51.8%)	17,845 (34.4%)	728 (30.5%)	1548 (17.5%)	240 (40.2%)	136 (24.2%)	2317 (48.5%)
Age, end of pregnancy (years)									<0.001
≤18	3348 (4.5%)	85 (1.4%)	2560 (4.9%)	69 (2.9%)	408 (4.6%)	15 (2.5%)	22 (3.9%)	189 (4.0%)
19–24	20,981 (28.0%)	909 (15.4%)	15,292 (29.5%)	664 (27.8%)	2523 (28.4%)	129 (21.6%)	151 (26.8%)	1313 (27.5%)
25–34	38,798 (51.7%)	3636 (61.5%)	26,213 (50.5%)	1259 (52.8%)	4427 (49.9%)	331 (55.4%)	316 (56.1%)	2616 (54.8%)
≥35	11,853 (15.8%)	1279 (21.6%)	7817 (15.1%)	393 (16.5%)	1513 (17.1%)	122 (20.4%)	74 (13.1%)	655 (13.7%)

Note: “Other” race/ethnicity includes American Indians, Alaska Natives, multiple race, “other” race or ethnicity, or unknown race/ethnicity. Birth order represents the births in the PROMISE dataset.

Abbreviations: AA/PI, Asian American/Pacific Islander; FPL, federal poverty level.

^a Row percentages.

^b Chi‐square test used for categorical variables; z‐test used for proportions; one‐way analysis of variance used for continuous variables.

FIGURE 1.

Sankey diagram representing the proportion of insurance changes among 74,980 pregnancies in the PROMISE study. This figure includes those with a single insurance change (i.e., those listed under “Experienced a single insurance change” in Table 3). Those in the “More Complex Pattern” designation include those with changes in more than one direction (i.e., those listed under “Both” for “Direction of Discontinuity” in Table 3) and those in the all‐3‐types pattern.

Racial and ethnic proportions varied across insurance patterns (p < 0.001). The continuous private group was predominantly comprised of White individuals (62.2%), while the continuous public and uninsured groups were predominantly comprised of Hispanic individuals (60.2% and 56.9%, respectively). Hispanic individuals also accounted for large percentages of the private/public and uninsured/public discontinuous groups (49.2% and 67.9%, respectively). The continuous private and uninsured groups had low percentages of Black individuals (7.9% and 3.1%, respectively), while the private/public (14.3%) and all‐3‐types (18.5%) discontinuity patterns had the largest percentages of Black individuals. The private/public group had the largest percentage of AA/PI individuals (7.3%), followed by the continuous public group (5.0%).

Individuals with continuous private insurance and uninsured/private discontinuity were most likely to be in the upper‐income range (19.6% and 18.1% had income >198% FPL, respectively), as compared to all other groups (between 2.1% and 8.4%), though the large percent missing FPL data (34.5%) precludes a thorough interpretation (p < 0.001). Maternal age also differed between insurance categories (p < 0.001), with the highest frequency of adolescent parents in the continuous public insurance category (4.9% ≤ 18 years old), followed by public/uninsured discontinuity (4.6%) and all‐3‐types discontinuity (3.9%). Advanced maternal age was most common in the continuous private (21.6% ≥ 35 years old) and the uninsured/private discontinuity groups (20.4%).

3.2. Health care utilization and insurance trajectories

Those with continuous public insurance had the highest frequency of inadequate PNC (19.5%), followed by the uninsured/public discontinuity group (15.9%) (Table 2; p < 0.001). The all‐3‐types discontinuity group had the highest percentage of intensive PNC (16.5%) followed by the private/public discontinuity group (15.2%), reflecting consistent engagement with clinical care and/or early initiation of PNC (90.9% and 88.6% initiated PNC in the first trimester, respectively; p < 0.001).

TABLE 2.

Intensity of health care utilization (N [column %]) among individuals in the PROMISE study, by insurance pattern (N = 74,980).

	Continuous		Discontinuous				Uninsured	p Value b
	Private	Public	Private/public	Uninsured/public	Uninsured/private	All‐3‐types
	N = 5909; 7.9% a	N = 51,882; 69.2% a	N = 2385; 3.2% a	N = 8871; 11.8% a	N = 597; 0.8% a	N = 563; 0.8% a	N = 4773; 6.4% a
Adequacy of PNC, N (%) c								<0.001
Intensive	378 (6.4%)	6739 (13.0%)	363 (15.2%)	968 (10.9%)	52 (8.7%)	93 (16.5%)	270 (5.7%)
Adequate	2842 (48.1%)	16,848 (32.5%)	1054 (44.2%)	3223 (36.3%)	259 (43.4%)	254 (45.1%)	1856 (38.9%)
Intermediate	2028 (34.3%)	18,190 (35.1%)	764 (32.0%)	3269 (36.9%)	222 (37.2%)	179 (31.8%)	2000 (41.9%)
Inadequate	661 (11.2%)	10,105 (19.5%)	204 (8.6%)	1411 (15.9%)	64 (10.7%)	37 (6.6%)	647 (13.6%)
Initiation of PNC in the first trimester, N (%)	4999 (84.6%)	40,661 (78.4%)	2114 (88.6%)	7751 (87.4%)	510 (85.4%)	512 (90.9%)	3729 (78.1%)	<0.001
Number of PNC visits, median (IQR)	12 (10–14)	11 (8–14)	13 (10–15)	11 (8–14)	12 (10–14)	13 (10–16)	11 (9–13)	0.42
Amount of time observed (weeks), mean (SD)	27.9 (7.5)	25.9 (8.1)	29.3 (6.5)	27.5 (7.3)	27.8 (7.1)	29.9 (6.3)	26.3 (8.0)	<0.001

Abbreviations: IQR, interquartile range; PNC, prenatal care; SD, standard deviation.

^a Row percentages.

^b Chi‐square test used for categorical variables; z‐test used for proportions; Kruskal–Wallis test used for medians; one‐way analysis of variance used for continuous variables.

^c Adequacy of PNC determined through the Revised‐Graduated Prenatal Care Utilization Index (Alexander & Kotelchuck, 1996).

Those with private/public or uninsured/public discontinuity predominantly switched to public insurance during pregnancy (54.5% and 74.8%, respectively; Table 3 and Figure 2), rather than switching from public insurance or oscillating between the two insurance types. Additionally, those with uninsured/public discontinuity spent ~4 times the amount of time on public insurance as they did uninsured, while those with private/public discontinuity spent only ~1.5 times the amount of time on public insurance as they did on private insurance.

TABLE 3.

Features of insurance discontinuity (N [column %]) among individuals of the PROMISE study who experienced discontinuity, by insurance pattern (N = 12,416; 16.6% of total population).

	Private/public	Uninsured/public	Uninsured/private	All‐3‐types
	N = 2385; 19.2% a	N = 8871; 71.4% a	N = 597; 4.8% a	N = 563; 4.5% a
Amount of time uninsured (weeks) b	‐	5.3 (8.3)	9.3 (10.8)	4.1 (7.7)
Amount of time with public (weeks) b	16.7 (10.7)	20.6 (9.6)	‐	13.9 (11.2)
Amount of time with private (weeks) b	11.9 (10.5)	‐	18.7 (10.6)	10.9 (10.8)
Number of visits uninsured b	‐	2.4 (2.4)	4.1 (3.9)	2.0 (2.0)
Number of visits with public b	7.3 (4.6)	8.9 (4.7)	‐	6.3 (4.6)
Number of visits with private b	5.4 (4.1)	‐	7.8 (4.2)	4.7 (3.9)
Direction of discontinuity c , d
Private to public	1299 (54.5%)	‐	‐	‐
Public to private	493 (20.7%)	‐	‐	‐
Both	593 (24.9%)	‐	‐	‐
If both, how many changes? b	3.2 (1.6)	‐	‐	‐
Uninsured to public	‐	6637 (74.8%)	‐	‐
Public to uninsured	‐	562 (6.3%)	‐	‐
Both	‐	1672 (18.8%)	‐	‐
If both, how many changes? b	‐	3.3 (1.4)	‐	‐
Uninsured to private	‐	‐	365 (61.1%)	‐
Private to uninsured	‐	‐	63 (10.6%)	‐
Both	‐	‐	169 (28.3%)	‐
If both, how many changes? b	‐	‐	3.5 (1.5)	‐
Experienced a single insurance change c	1792 (75.1%)	7199 (81.2%)	428 (71.7%)	‐
Experienced a more complex insurance trajectory c	593 (24.9%)	1672 (18.8%)	169 (28.3%)	563 (100%)
Number of discontinuities b	1.5 (1.1)	1.4 (1.1)	1.7 (1.4)	3.2 (1.6)

^a Row percentages.

^b Data presented as mean (standard deviation).

^c Data presented as N (%).

^d The direction of discontinuity is not shown for the all‐3‐types group because of complexity and small cell sizes.

FIGURE 2.

Graphs of insurance changes by gestational week and insurance discontinuity. Insurance type is represented by color: public (blue), private (red), and uninsured (green). Pregnancies are represented from their first visit until delivery. Insurance type is assumed to continue until an insurance change is seen at a visit.

Meanwhile, the highest prevalence of people switching to uninsurance was in the uninsured/private discontinuity group (10.6%). This group also had the highest prevalence of experiencing changes in both directions (e.g., from uninsured to private to uninsured; 28.3%).

The majority of those with a discontinuous pattern experienced a single insurance change, from 71.7% (uninsured/private) to 81.2% (uninsured/public).

3.3. Sensitivity analysis

When we re‐ran analyses without correcting for single‐visit inconsistencies, membership in the discontinuous patterns increased while membership in the continuous patterns decreased. The degree of increase ranged from a 13.8% increase in the private/public discontinuous pattern (N = 2385–2713) to a 46.6% increase in the uninsured/private discontinuous pattern (N = 597–875). The degree of decrease ranged from a 2.4% decrease in the continuous uninsured pattern (N = 4773–4658) to a 6.8% decrease in the continuous private pattern (N = 5909–5510). The all‐3‐types group did not change, as they necessarily did not have only one single‐visit inconsistency and so were not affected by this correction. A discontinuous pattern was seen in 20.8% of the total population, as compared to 16.6% when single‐visit inconsistencies were corrected.

All of the abovementioned relationships held when not correcting for single‐visit inconsistencies, though to varying degrees (see Supplementary Materials).

4. DISCUSSION

Using data from a large, EHR‐based cohort from clinics that treat people regardless of insurance status, this study has defined and demonstrated continuous and discontinuous insurance patterns during pregnancy. We found that insurance discontinuity was common, with at least 16.6% of our population experiencing discontinuity during their pregnancy (as high as 20.8%, depending on data cleaning). This represents a larger group of people than those with continuous private insurance (7.9%) or uninsurance (6.4%) combined. These findings confirm that around the time of pregnancy, insurance status should be regarded as a dynamic rather than a static feature and should be measured accordingly.

We observed differences in sociodemographic characteristics between insurance groups; for example, the uninsured/public discontinuous group was predominantly Hispanic, along with the continuously publicly insured and uninsured groups. In contrast, the continuous private insurance group and the uninsured/private discontinuous group were predominantly made up of White individuals.

Studies using cross‐sectional and self‐reported data have reported the presence of insurance discontinuities during pregnancy, ⁸ , ⁹ , ¹² , ¹³ , ¹⁹ and one study has used provider‐recorded longitudinal data to enumerate transitions in continuous insurance. ²⁰ Our study is the first to use provider‐recorded longitudinal data that includes uninsured individuals to detail the timing and nuances of insurance churn. We report a lower prevalence of insurance churn (16.6%) than previous work that used MEPS‐HC (58%) ⁸ or PRAMS (30.1% ⁹ and 48.7% ¹⁹ ) data. There are several likely explanations for these differences. First, while MEPS‐HC and PRAMS utilized population‐representative samples, we studied insurance trajectories among pregnant women receiving PNC at CHCOs, many of which were FQHCs. These individuals bear a disproportionate share of medical risk, are under‐represented in most research, and are more likely than the general population to qualify for regular Medicaid (due to income). The higher percentage of continuously publicly insured people in our study (69.2%) probably reflects this.

Second, our study has notably different data collection and provenance compared to prior studies of perinatal insurance churn. For example, MEPS‐HC and PRAMS contain self‐reported insurance data from a cross‐sectional sampling frame, often collected at timepoints outside of pregnancy (e.g., in the postpartum period for PRAMS). Given the complexity of health insurance in the US (especially surrounding pregnancy), retrospective recall of insurance status presents challenges. In contrast, this study used repeated measures of the payer billed at the time of care delivery and/or payment, which removes issues of self‐report and historical recall, and also provides longitudinal data to directly document insurance discontinuity. Perhaps for this reason, our estimate of insurance churn is closer in magnitude to findings from a recent study (14.4%) using the Massachusetts All‐Payer Claims Database to classify insurance transitions in the 12 months before delivery among those with continuous insurance (e.g., transitions between private insurance and Medicaid). ²⁰

Demographic differences between insurance trajectories are complex and might relate to the numerous factors that can cause someone to shift between insurance (due to choice, external circumstances, or both). One such consideration is the income threshold for Medicaid eligibility, which is complex and varies according to state, pregnancy status, and documentation status. The income eligibility threshold for pregnancy‐related Medicaid coverage is higher (e.g., 213% FPL in California and 190% FPL in Oregon ¹⁵ ) than the threshold for non‐pregnancy‐related Medicaid (e.g., 138% FPL for both California and Oregon ³⁵ ), so some women qualify for Medicaid when they become pregnant who would not otherwise qualify. The largest single insurance discontinuity type observed in our population was a switch from uninsured to public insurance (N = 6637; 53.5% of all discontinuities and 8.9% of the entire sample); this in part might reflect women seeking to obtain insurance that is not otherwise available (e.g., due to lack of employer‐sponsored health insurance and/or lack of other affordable options).

This expanded eligibility during pregnancy likely also explains the next most common single insurance discontinuity type we observed: a switch from private to public insurance (N = 1299; 10.5% of all discontinuities and 1.7% of our overall population). There is no cost‐sharing for perinatal care in Medicaid, and these expenses (e.g., co‐pays, deductibles, and co‐insurance fees) can be substantial for pregnant people with private health insurance ($4569 average out‐of‐pocket cost in 2015). ³⁶ Other life changes that can precipitate perinatal changes in health insurance status include changes in employment status (relevant to employer‐based private insurance) and marriage or divorce (relevant to qualifying as a dependent on a spouse's insurance plan). Collectively, the factors that affect a pregnant person's insurance status and continuity are complex and interact in various ways that lead to widespread insurance discontinuities. Although our data did not permit us to ascertain the drivers of insurance changes, future research should explore and document these to better understand the causes of churn.

Finally, we observed a moderately high percentage of uninsurance, with 6.4% of our study population remaining uninsured for the entirety of their pregnancy. This is larger than the national rate (3.9% self‐paid for childbirth in 2021 ¹⁸ ), likely due to our unique study setting within CHCOs. These individuals had among the highest percentage of inadequate PNC, the lowest percentage initiating care in the first trimester, and among the shortest amount of time observed. Notably, one population that is ineligible for prenatal Medicaid coverage regardless of their income is undocumented immigrants (with exceptions). ³⁷ These individuals are expected to remain uninsured throughout the prenatal period until childbirth hospital admission and/or the development of an acute life‐threatening health condition, when they become eligible for Emergency Medicaid. ³⁷ Additionally, among those who started their pregnancy insured, becoming uninsured was not uncommon (5.0% of those with a discontinuous insurance pattern). It will be important in future research to document any differing perinatal outcomes and to examine the effects of less restrictive insurance policies on care utilization and health outcomes among those who lose or never gain insurance during pregnancy.

We compared health care utilization between insurance categories, with the acknowledgment that PNC utilization and designation of insurance trajectory are inter‐related in this data source (i.e., more utilization provides more opportunities to detect insurance discontinuity). Because of this feature of our data, we caution that differences in utilization cannot necessarily be interpreted as resulting from the insurance trajectory. Additionally, because of our inclusion criterion necessitating at least three visits, it is possible that inadequate PNC is underrepresented. With these caveats, we found that those with continuous public insurance had the highest proportion of inadequate PNC utilization, indicating either late entry into PNC (possibly due to the need to enroll in public insurance) and/or fewer clinical visits attended. These findings reflect the complexity of prenatal health insurance eligibility and care utilization in the United States. It is also possible that some proportion of those whom we have categorized as “continuous public insurance” actually enrolled in public insurance while pregnant but before their first prenatal visit, in which case we would have erroneously recorded them as having continuous public insurance.

We have shed light on the frequency and nature of perinatal health insurance trajectories using longitudinal data, but it will be equally important to assess any impacts of these discontinuities on health care utilization and perinatal outcomes. For example, insurance disruptions may delay care and reduce the number of PNC visits that pregnant people attend (especially in the case of uninsurance but potentially even if there is no gap in coverage). ²⁴

Our study has many strengths, including the incorporation of uninsured visits and the thorough detail of each insurance pattern. We included EHR data from 855,426 visits from 74,980 pregnancies among 62,885 pregnant people who sought care at CHCOs. This large sample provided us with the nuance to describe patterns related to demographics and health care utilization within insurance pattern groups and the timing and direction of discontinuity. Our data was not complicated by issues of self‐report or historical recall given that it was provider‐recorded at the time of care delivery or payment.

Our study is not without limitations. We do not report transitions between specific plans within each insurance type (e.g., different private insurance plans); while this was not the purpose of this analysis, it is possible that transitions within the same insurance type could still elicit insurance disruptions. Additionally, interpretation regarding FPL is limited due to substantial missingness. It is additionally possible that insurance changes may trigger a transfer of care to or from a CHCO. For example, becoming uninsured may elicit a transfer to a CHCO while gaining insurance may elicit a transfer out of a CHCO. Such clinic transfers (and other factors) may affect our results. For example, insurance discontinuities may be underrepresented in our analysis or prenatal visits may be undercounted (in some cases potentially leading to exclusion from the study sample). Further, our study period spans many years (2005–2021) and, during this time, policies were enacted that directly affected health insurance accessibility in the US (e.g., the Affordable Care Act). While it was not the purpose of this analysis to assess time trends, the prevalence and patterns of insurance discontinuities may have changed over time with the changing political climate. Finally, the PROMISE study does not currently have access to individual‐level state of residence at the time of pregnancy, so while we can share aggregate counts of patients by state, we were unable to assess associations between state‐level policies and insurance patterns.

5. CONCLUSIONS

In this study of longitudinal data with contemporaneous recording of insurance status, we found that insurance discontinuity is common through the prenatal period, with a prevalence larger than the prevalence of continuous private insurance and uninsurance combined. Although the drivers behind changes in insurance status could not be captured in this study, the types of insurance changes that were most common in our population align with what is known about the reasons for changing insurance status (including affordability and access). Given the widespread nature of insurance discontinuities in the prenatal period, future research should consider insurance status during pregnancy as a dynamic rather than a static characteristic. Finally, research is needed to assess if and how these prenatal insurance discontinuities affect health care utilization and perinatal health outcomes.

CONFLICT OF INTEREST STATEMENT

All authors report no competing interests.

Supporting information

Data S1. Supplementary information.

Booman A, Stratton K, Vesco KK, et al. Insurance coverage and discontinuity during pregnancy: Frequency and associations documented in the PROMISE cohort. Health Serv Res. 2024;59(2):e14265. doi: 10.1111/1475-6773.14265