Duration of virtual maternity program engagement and neonatal outcomes: A multistate retrospective cohort study

Abstract

Introduction

Virtual maternity solutions vary widely in clinical depth, and their association with physiological outcomes in high‐risk populations remains understudied. We evaluated the association between the duration of engagement in a multispecialty virtual maternity program and neonatal outcomes among a medically complex population.

Methods

We conducted a retrospective cohort study using administrative claims from six Medicaid‐managed care plans across six US states (January 2024–April 2025). We compared outcomes of patients enrolled in a virtual program, integrating remote patient monitoring, multispecialty care, and social support, against propensity‐weighted comparators receiving standard care. The population exhibited a high comorbidity burden, including 36% rates of hypertension and 50% of obesity. The primary exposure was engagement duration (≥1, ≥2, and ≥3 months). Primary outcomes were preterm birth, neonatal intensive care unit (NICU) admission, and NICU length of stay (LOS), estimated using doubly robust modified Poisson regression with inverse probability of treatment weighting (IPTW).

Results

The analysis of longer engagement (≥3 months) included 3785 treated patients and 38,180 weighted comparators. Engagement of ≥3 months was associated with a 24.4% reduction in the adjusted relative risk of preterm birth (95% CI, −32.5% to −15.4%; p < 0.001), decreasing the absolute rate from 11.6% to 8.8%. This cohort also experienced a 12.5% reduction in NICU admissions (95% CI, −21.0% to −3.2%; p = 0.01) and a 26.3% reduction in NICU LOS (95% CI, −38.0% to −12.4%; p < 0.001). Shorter engagement durations were associated with smaller reductions, with effect sizes increasing with longer engagement. These findings remained significant in a sensitivity analysis restricted to participants enrolled prior to 28 weeks’ gestation, mitigating time‐dependent bias.

Conclusion

Longer participation in a clinically integrated virtual maternity program is associated with duration‐correlated reductions in preterm birth and NICU utilization among high‐risk beneficiaries. Virtual models capable of maintaining longitudinal engagement may offer clinical value in managing complex obstetric populations.

1. INTRODUCTION

The digital health landscape has expanded to offer a diverse array of virtual maternity solutions. These platforms are increasingly viewed as essential tools to address the structural barriers, such as transportation challenges, work schedule conflicts, and geographic distance, which disproportionately affect vulnerable populations and impede consistent engagement with traditional care models [1, 2, 3]. However, significant heterogeneity exists in the design and clinical depth of these models. Most widely deployed solutions include educational content (e.g., an article library), peer support, care coordination, or coaching [4]. These interventions have been shown to improve patient satisfaction, but their capacity to alter the trajectory of complex physiological pathologies (e.g., adverse pregnancy outcomes) is theoretically limited [5, 6]. Effective management of high‐risk obstetrical conditions generally requires protocol‐driven medical management, longitudinal clinical relationships, and the ability to integrate directly with the medical record and broader medical ecosystem, capabilities that may be absent in platforms relying on ad hoc or nonclinical provider networks [7, 8].

This gap in clinical capability is significant, as preterm birth affects approximately 10% of all births in the United States and remains a leading cause of neonatal morbidity and mortality [9]. The clinical and economic consequences are substantial, with annual societal excess costs associated with prematurity over $25 billion, driven primarily by prolonged neonatal intensive care unit (NICU) admissions [10]. These burdens are not borne equally; Medicaid beneficiaries and racially and ethnically minority populations, particularly non‐Hispanic Black individuals, as well as those living in rural settings, experience significantly higher rates of preterm birth and adverse outcomes than the general population [11, 12].

Traditional prenatal care models, which rely on periodic in‐person visits, are rapidly becoming insufficient for the volume and acuity of prenatal care required by pregnant people in the United States today [13]. The episodic nature of scheduled appointments means that complications evolving between visits (e.g., progressive hypertension, worsening glycemic control, escalating mental health symptoms) may go undetected until patients present acutely [13]. In‐person care is typically unavailable during evenings, nights, and weekends when many pregnancy symptoms emerge, leaving patients without clinical guidance during critical windows.

Furthermore, the potential for virtual care to reduce disparities is often constrained by enrollment mechanisms. Programs that rely on provider referrals or an existing diagnosis inherently reach and benefit only those patients already successfully connected to the prenatal care system, potentially widening the gap for high‐risk individuals who are unattached to care or face significant access barriers [14]. To achieve population‐level improvements and close the health equity gap in maternity, interventions must possess both the clinical rigor to manage medical complexity and the infrastructure to proactively identify and engage vulnerable populations independent of traditional referral pathways [14, 15].

Finally, while the concept of “adequacy of prenatal care” is well‐established in traditional settings, the relationship between the duration of virtual care and outcomes remains less defined. A recent systematic review by the Agency for Healthcare Research and Quality (AHRQ) highlighted the heterogeneity of virtual care models and the need for evidence regarding optimal modalities [16].

The objective of this study was to evaluate the association between the duration of engagement with a virtual maternity program and neonatal outcomes, specifically NICU utilization and preterm birth rates, among a diverse Medicaid population.

2. METHODS

2.1. Study design and population

We conducted a retrospective cohort study using administrative claims and enrollment data from six Medicaid‐managed care plans across six US states (Texas, Georgia, Nevada, New York, Louisiana, and Missouri) covering deliveries between January 1, 2024, and April 30, 2025. Data sources included member rosters, International Classification of Diseases 10 (ICD‐10) diagnosis codes, Current Procedural Terminology (CPT) procedure codes, and revenue codes for facility stays to capture the full granularity of clinical utilization. The study compared neonatal outcomes among patients enrolled in a virtual maternity program with varying engagement durations to propensity‐weighted comparators receiving standard prenatal care. The virtual maternity program launched sequentially across markets between January and September 2024, creating natural variation in patient opportunity for enrollment and sustained engagement.

This study was conducted and reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for cohort studies.

2.2. Inclusion and exclusion criteria

Eligible participants included Medicaid beneficiaries with singleton or multiple gestations who delivered at least one liveborn infant during the study period. Deliveries with missing infant claims linkage were excluded. For regression analyses, we applied additional exclusions to ensure complete covariate data and remove extreme cost outliers. We excluded deliveries with missing data for any covariate used in propensity models (race/ethnicity, maternal age, insurance start date, first prenatal visit date, or other risk adjusters). Deliveries with total delivery costs in the bottom first percentile by market were also excluded to address data quality issues. A study flow diagram details the attrition from the raw population to the final cohort (Figure 1).

FIGURE 1.

Cohort flow diagram.

2.3. Intervention: Virtual maternity program

The intervention was a virtual maternity program designed to integrate medical management, behavioral and social care, and care coordination. Distinct from referral‐based models that rely on established prenatal care connections, this program utilized health plan data to proactively identify, enroll, and engage eligible beneficiaries, including those unattached to traditional care at enrollment.

Patients accessed a multispecialty prenatal team (obstetricians, midwives, nurses, dietitians, lactation consultants, and social workers) via virtual modalities including a mobile application, SMS, phone, and video visits. The model featured 24/7 asynchronous clinical access with escalation to synchronous visits. Beyond educational content and coaching, the care team provided medical and behavioral care and population health services including conducting proactive outreach and utilizing evidence‐based protocols to actively manage pregnancy‐related conditions. Remote patient monitoring (e.g., blood pressure cuffs, glucometers) allowed clinicians to review physiologic data in real‐time and adjust medications or dietary plans dynamically between in‐person visits.

Behavioral healthcare was embedded within the primary care team. Universal screening for mood and anxiety disorders was followed by immediate “warm handoffs” to internal therapists and behavioral health specialists, eliminating the attrition common in traditional referral workflows. Registered nurses also conducted systematic screening for social determinants of health, facilitating direct linkage to community resources. For infants requiring intensive care, a specialized virtual NICU team provided longitudinal family support, discharge planning, and virtual care after discharge. These individual clinical services were complemented by opt‐in virtual group prenatal care sessions and tailored educational resources. The virtual model was designed to wrap around, not replace, in‐person obstetric care, with clear protocols for conditions requiring in‐person evaluation.

2.4. Exposure definition

The primary exposure was engagement duration, operationalized as the number of days between first program engagement and delivery. Engagement was defined as active, bi‐directional interaction (messaging, video, or phone) with the clinical team. We created three mutually inclusive exposure categories: ≥1 month (≥30 days), ≥2 months (≥60 days), and ≥3 months (≥90 days) of prenatal engagement. These cutoffs were selected to represent meaningful intervals of prenatal care delivery, mirroring the periodicity of standard prenatal visits. By definition, all treated individuals first engaged at least 30 days before delivery.

2.5. Covariates

We extracted baseline covariates from claims and enrollment data. Demographic characteristics included maternal age, self‐reported race and ethnicity, and urban versus rural residence. Race and ethnicity were utilized as sociodemographic variables to assess population representativeness and equitable program delivery. Insurance history variables captured length of claims history prepregnancy and timing of Medicaid enrollment relative to pregnancy (categorized as before pregnancy, first trimester, second trimester, or third trimester). Timing of first prenatal visit was similarly categorized by trimester. Healthcare utilization measures included counts of emergency department and inpatient encounters in the period prior to delivery. Race and ethnicity were self‐reported by the beneficiary at the time of Medicaid enrollment.

Pre‐existing and gestational medical conditions were identified through diagnosis codes appearing any time before the delivery admission date. This approach captured both chronic conditions diagnosed before pregnancy and gestational conditions (e.g., gestational diabetes, pregnancy‐related hypertension) diagnosed during prenatal care, which was important given variable lengths of claims history in this Medicaid population. Conditions included hypertension (chronic and pregnancy‐related), diabetes (pregestational and gestational), prediabetes, obesity, overweight, nicotine use disorder, opioid use disorder, alcohol use disorder, perinatal mood and anxiety disorders, and severe mental health diagnoses. Obstetric history included prior cesarean delivery. Gestational age (GA) at program enrollment (for treated) and roster/index date (for comparators) was used as a covariate in the propensity models. We also adjusted for temporal and geographic trends by including delivery quarter (year and quarter of delivery) and health plan.

2.6. Outcome measures

Primary outcomes were ascertained from administrative claims. Preterm birth was defined as delivery at fewer than 37 completed weeks of gestation, identified using diagnosis codes, procedure codes, and gestational age fields. NICU admission was defined as any NICU stay for a liveborn infant, identified through revenue codes and place of service codes. NICU length of stay was defined as mean days from NICU admission to discharge across all deliveries (including 0 days for infants without NICU admission).

2.7. Propensity score methods

We developed separate propensity score models using logistic regression for each engagement duration threshold (≥1 month, ≥2 months, and ≥3 months). We implemented inverse probability of treatment weighting (IPTW) using stabilized weights to create weighted pseudo‐populations where covariate distributions were balanced between exposed and unexposed groups. We used the average treatment effect (ATE) estimand to maintain larger effective sample sizes. Weights were truncated at the first and 99th percentiles to reduce variance inflation from extreme weights. Covariate balance was assessed for each engagement cohort using standardized mean differences (SMDs).

2.8. Outcome analysis

Outcomes were estimated using doubly robust regression combining inverse probability weighting with covariate adjustment. In this approach, all covariates used in the propensity score model (including gestational age at enrollment) were also included as independent variables in the final outcome models to adjust for any residual imbalance remaining after weighting.

For binary outcomes (preterm birth, NICU admission), we used weighted modified Poisson regression with robust error variance to estimate relative risks (RR). For NICU length of stay, we used weighted Poisson regression with a log link. Percentage changes were calculated by exponentiating regression coefficients, with 95% confidence intervals derived from robust standard errors.

To assess engagement duration relationships, we examined the pattern and magnitude of effect estimates across the three engagement duration thresholds. Two‐sided p‐values less than 0.05 were considered statistically significant. We did not adjust for multiple comparisons given the hierarchical nature of the exposure categories and the a priori specification of engagement duration assessment as the primary analytical framework.

2.9. Sensitivity analysis

To account for potential confounding related to the opportunity for engagement, we conducted a sensitivity analysis restricted to the subset of the population enrolled or rostered prior to 28 weeks’ gestation. Within this restricted cohort, in which all individuals possessed the theoretical temporal opportunity to achieve the maximal engagement duration of ≥3 months, we utilized a multivariable modified Poisson regression model, adjusted for all primary covariates, to estimate the Relative Risk of preterm birth.

3. RESULTS

3.1. Study population

The final analytic cohort for the ≥1 month comparison included 6836 treated patients and 38,180 comparators (N = 45,016). For the ≥2 month comparison, 5257 treated patients were included, and for the primary ≥3 month analysis, 3785 treated patients were analyzed against the same comparator pool (Table 4). Mean gestational age at delivery was 38.1 weeks (SD 2.4) for the comparator group and 38.3 weeks (SD 2.1) for the high‐engagement (≥3 months) cohort.

TABLE 4.

Neonatal outcomes by engagement duration.

Outcome	Duration	N engaged	N comparator	Comparator rate/mean	Engaged rate/ mean	% Change (95% CI)	p value
Preterm birth (%)	≥1 month	6836	38,180	11.6%	10.40%	−11.0% (−17.8 to −3.5)	<0.01 **
	≥2 months	5257	38,179	11.6%	9.70%	−17.2% (−24.6 to −9.1)	<0.001 ***
	≥3 months	3785	38,180	11.6%	8.80%	−24.4% (−32.5 to −15.4)	<0.001 ***
NICU admissions (%)	≥1 month	6836	38,180	13.2%	12.40%	−6.0% (−12.8 to 1.4)	0.11
	≥2 months	5257	38,179	13.2%	12.20%	−7.5% (−15.1 to 0.7)	0.07 †
	≥3 months	3785	38,180	13.2%	11.60%	−12.5% (−21.0 to −3.2)	0.01 *
NICU length of stay (days)	≥1 month	6836	38,180	2.00	1.69	−16.5% (−26.7 to −5.0)	0.01 *
	≥2 months	5257	38,179	2.00	1.67	−17.1% (−28.6 to −3.8)	0.01 *
	≥3 months	3785	38,180	2.00	1.49	−26.3% (−38.0 to −12.4)	<0.001 ***

Abbreviations: CI, confidence interval; NICU, neonatal intensive care unit.

^† p < 0.10.

^* p < 0.05.

^** p < 0.01.

^*** p < 0.001.

3.2. Propensity weighting

Before weighting, enrolled patients had higher rates of conditions associated with pregnancy complications including hypertension (36.1% vs. 25.7%), obesity (50.0% vs. 41.0%), and diabetes (18.4% vs. 15.2%). They were more likely to have enrolled in Medicaid during pregnancy (35.8% vs. 28.2%). After weighting, robust covariate balance was achieved for most demographic and clinical history variables. However, residual imbalance (SMD > 0.10) persisted for gestational age at enrollment (SMD 0.42), as well as timing of insurance enrollment (SMD 0.11) and timing of first prenatal visit (SMD 0.11). These temporal covariates were explicitly adjusted for in the final doubly robust outcome models to mitigate potential bias. Tables 1, 2, 3 present baseline characteristics and balance for all cohorts. The improvement in covariate balance after weighting is visualized in Figure 2 for the longest‐exposure cohort (≥ 3 months).

TABLE 1.

Baseline characteristics of longest exposure cohort (≥3 months) before/after weighting.

	Before weighting			After weighting
Characteristic	Weighted comparator	Engaged ≥3 months	SMD	Weighted comparator	Engaged ≥3 months	SMD
Demographics
Maternal age, mean (SD), years	26.9 (5.9)	28.1 (5.8)	0.20	27.0 (5.9)	27.3 (5.7)	0.05
Race/ethnicity, %			0.15			0.05
White, non‐Hispanic	36.4	30.5		35.8	34.9
Black, non‐Hispanic	27.7	33.3		28.2	29.7
Hispanic/Latinx	19.3	17.8		19.1	17.9
Asian	2.7	3.3		2.8	2.7
American Indian/Alaska Native	0.4	0.4		0.4	0.4
Native Hawaiian/Pacific Islander	0.3	0.3		0.3	0.3
Other/unknown	13.3	14.3		13.4	14.2
Rural residence, %	23.3	18.7	0.11	22.9	22.2	0.02
Insurance and care access
Length of claims history prepregnancy, mean (SD), days	399 (421)	375 (436)	0.06	397 (422)	427 (435)	0.07
Gestational age at enrollment, mean (SD), weeks	15.9 (11.3)	9.4 (7.9)	0.66	15.3 (11.3)	11.1 (8.2)	0.42
Insurance start timing, %			0.30			0.11
Before pregnancy	71.8	64.3		71.1	71.8
First trimester	13.7	21.6		14.3	15.2
Second trimester	8.7	12.1		9	9.7
Third trimester	5.2	1.8		4.9	3
Unknown	0.6	0.2		0.6	0.2
First prenatal visit timing, %			0.24			0.11
First trimester	46.4	51.9		46.9	49
Second trimester	28.1	31.9		28.5	30.8
Third trimester	17	9.8		16.3	13.1
No documented prenatal visit	8.5	6.4		8.3	7.1
Prepregnancy healthcare utilization
Emergency department visits, mean (SD)	2.1 (3.2)	2.2 (3.3)	0.06	2.1 (3.2)	2.3 (3.3)	0.06
Inpatient visits, mean (SD)	0.09 (0.77)	0.12 (1.19)	0.03	0.09 (0.78)	0.09 (0.96)	0.00
Medical comorbidities, %
Hypertension	25.7	36.1	0.23	26.7	30.7	0.09
Diabetes	15.2	18.4	0.09	15.5	17	0.04
Prediabetes	11.1	14.2	0.09	11.4	12.8	0.04
Obesity	41	50	0.18	41.9	46.1	0.09
Overweight	6.8	7.6	0.03	6.9	7.4	0.02
Substance use disorders, %
Nicotine use disorder	13.5	11.9	0.05	13.4	13.9	0.02
Opioid use disorder	1.5	0.9	0.06	1.4	1.3	0.01
Alcohol use disorder	2	1.6	0.03	1.9	2	0.00
Mental health, %
Any mental health diagnosis	28.6	31.3	0.06	28.9	31.6	0.06
Severe mental health diagnosis	0.6	0.7	0.00	0.7	0.8	0.02
Obstetric history, %
Prior cesarean delivery	8	7.4	0.02	7.9	8.3	0.01

Abbreviation: SMD, standardized mean difference.

TABLE 2.

Baseline characteristics of middle‐exposure cohort (≥2 months) before/after weighting.

	Before weighting			After weighting
Characteristic	Weighted comparator	Engaged ≥2 months	SMD	Weighted comparator	Engaged ≥2 months	SMD
Demographics
Maternal age, mean (SD), years	26.9 (5.9)	27.9 (5.8)	0.17	27.0 (5.9)	27.2 (5.7)	0.03
Race/ethnicity, %			0.14			0.04
White, non‐Hispanic	36.3	31.1		35.7	35.1
Black, non‐Hispanic	27.7	33.0		28.3	29.6
Hispanic/Latinx	19.3	18.1		19.1	17.9
Asian	2.7	3.0		2.8	2.7
American Indian/Alaska Native	0.4	0.4		0.4	0.4
Native Hawaiian/Pacific Islander	0.3	0.3		0.3	0.3
Other/unknown	13.3	14.1		13.4	14.1
Rural residence, %	23.3	18.9	0.11	22.8	22.3	0.01
Insurance and care access
Length of claims history prepregnancy, mean (SD), days	399 (421)	374 (435)	0.06	396 (423)	422 (433)	0.06
Gestational age at enrollment, mean (SD), weeks	15.9 (11.3)	10.1 (8.4)	0.58	15.2 (11.3)	12.0 (8.7)	0.31
Insurance start timing, %			0.26			0.07
Before pregnancy	71.8	64.7		71.0	71.5
First trimester	13.7	20.1		14.4	14.7
Second trimester	8.7	12.3		9.2	9.5
Third trimester	5.2	2.7		4.9	4.1
Unknown	0.6	0.2		0.5	0.2
First prenatal visit timing, %			0.19			0.06
First trimester	46.4	50.5		46.9	48.1
Second trimester	28.1	31.8		28.6	29.9
Third trimester	17.0	11.4		16.3	14.6
No documented prenatal visit	8.5	6.4		8.2	7.4
Prepregnancy healthcare utilization
Emergency department visits, mean (SD)	2.1 (3.2)	2.2 (3.2)	0.05	2.1 (3.2)	2.2 (3.1)	0.04
Inpatient visits, mean (SD)	0.09 (0.77)	0.11 (1.11)	0.03	0.09 (0.79)	0.09 (0.92)	0.00
Medical comorbidities, %
Hypertension	25.7	34.4	0.19	26.8	29.1	0.05
Diabetes	15.2	18.4	0.08	15.6	16.6	0.03
Prediabetes	11.1	14.1	0.09	11.5	12.4	0.03
Obesity	41.0	49.0	0.16	42.0	44.5	0.05
Overweight	6.8	7.6	0.03	6.9	7.6	0.03
Substance use disorders, %
Nicotine use disorder	13.5	12.1	0.04	13.3	13.5	0.00
Opioid use disorder	1.5	0.9	0.05	1.4	1.3	0.01
Alcohol use disorder	2.0	1.8	0.01	2.0	2.1	0.01
Mental health, %
Any mental health diagnosis	28.6	30.6	0.04	28.9	30.8	0.04
Severe mental health diagnosis	0.6	0.6	0.01	0.6	0.7	0.01
Obstetric history, %
Prior cesarean delivery	8.0	7.3	0.03	7.9	8.1	0.01

Abbreviation: SMD, standardized mean difference.

TABLE 3.

Baseline characteristics of shortest‐exposure cohort (≥1 months) before/after weighting.

	Before weighting			After weighting
Characteristic	Weighted comparator	Engaged ≥1 month	SMD	Weighted comparator	Engaged ≥1 month	SMD
Demographics
Maternal age, mean (SD), years	26.9 (5.9)	27.6 (5.8)	0.13	27.0 (5.9)	27.1 (5.6)	0.02
Race/ethnicity, %			0.10			0.03
White, non‐Hispanic	36.3	32.5		35.8	35.2
Black, non‐Hispanic	27.7	31.7		28.3	29.3
Hispanic/Latinx	19.3	18.5		19.1	18.2
Asian	2.7	2.8		2.7	2.6
American Indian/Alaska Native	0.4	0.5		0.4	0.4
Native Hawaiian/Pacific Islander	0.3	0.4		0.3	0.3
Other/unknown	13.3	13.8		13.4	14
Rural residence, %	23.3	19.7	0.09	22.7	22.4	0.01
Insurance and care access
Length of claims history prepregnancy, mean (SD), days	399 (421)	380 (436)	0.04	397 (424)	421 (432)	0.06
Gestational age at enrollment, mean (SD), weeks	15.9 (11.3)	10.1 (8.4)	0.58	15.0 (11.3)	12.1 (8.8)	0.28
Insurance start timing, %			0.23			0.06
Before pregnancy	71.8	65.9		71.0	71.6
First trimester	13.7	19.2		14.4	14.6
Second trimester	8.7	11.9		9.2	9.4
Third trimester	5.2	2.8		4.9	4.2
Unknown	0.6	0.2		0.5	0.3
First prenatal visit timing, %			0.20			0.05
First trimester	46.4	50.2		47.0	48.1
Second trimester	28.1	32.1		28.8	29.8
Third trimester	17.0	11.2		16.1	14.4
No documented prenatal visit	8.5	6.4		8.2	7.6
Prepregnancy healthcare utilization
Emergency department visits, mean (SD)	2.1 (3.2)	2.2 (3.2)	0.04	2.1 (3.2)	2.2 (3.1)	0.03
Inpatient visits, mean (SD)	0.09 (0.77)	0.11 (1.02)	0.02	0.09 (0.78)	0.09 (0.88)	0.00
Medical comorbidities, %
Hypertension	25.7	32.8	0.15	26.8	28.5	0.04
Diabetes	15.2	17.8	0.07	15.6	16.3	0.02
Prediabetes	11.1	13.3	0.07	11.5	12.1	0.02
Obesity	41.0	48.1	0.14	42.2	44.3	0.04
Overweight	6.8	7.3	0.02	6.9	7.4	0.02
Substance use disorders, %
Nicotine use disorder	13.5	12.4	0.03	13.3	13.5	0.00
Opioid use disorder	1.5	1.1	0.04	1.4	1.4	0.00
Alcohol use disorder	2.0	1.8	0.01	1.9	2.1	0.01
Mental health, %
Any mental health diagnosis	28.6	29.8	0.03	28.9	30.5	0.04
Severe mental health diagnosis	0.6	0.6	0.01	0.6	0.8	0.01
Obstetric history, %
Prior cesarean delivery	8.0	7.6	0.01	7.9	8.2	0.01

Abbreviation: SMD, standardized mean difference.

FIGURE 2.

Covariate balance before and after weighting (Love plot) of SMD, standardized mean difference.

3.3. Neonatal outcomes: Engagement duration effects

Table 4 and Figure 3 present outcomes across three engagement durations, demonstrating engagement duration relationships for preterm birth and NICU outcomes.

FIGURE 3.

Forest plot showing engagement duration pattern NICU, neonatal intensive care unit.

Preterm birth. Among patients engaged ≥1 month, the adjusted risk of preterm birth declined 11.0% (95% CI, −17.8% to −3.5%; p < 0.01). At ≥2 months, the reduction increased to 17.2% (95% CI, −24.6% to −9.1%; p < 0.001). Patients engaged ≥3 months experienced a 24.4% reduction (95% CI, −32.5% to −15.4%; p < 0.001), representing an absolute decrease from 11.6% to 8.8% (2.8 percentage points).

NICU admission. We observed progressive reductions in NICU admission rates. At ≥1 month: 6.0% reduction (95% CI, −12.8% to 1.4%; p = 0.11). At ≥2 months: 7.5% reduction (95% CI, −15.1% to 0.7%; p = 0.07). At ≥3 months: 12.5% reduction (95% CI, −21.0% to −3.2%; p = 0.01), representing an absolute decrease from 13.2% to 11.6% (1.6 percentage points).

NICU length of stay. All engagement durations showed reductions in NICU days. At ≥1 month: 16.5% reduction (95% CI, −26.7% to −5.0%; p = 0.01). At ≥2 months: 17.1% reduction (95% CI, −28.6% to −3.8%; p = 0.01). At ≥3 months: 26.3% reduction (95% CI, −38.0% to −12.4%; p < 0.001), representing a decrease from 2.00 to 1.49 days (0.51 days).

3.4. Sensitivity analysis

In the restricted sub‐analysis of patients enrolled prior to 28 weeks’ gestation (N = 41,133), the association between high‐duration engagement (≥3 months) and reduced preterm birth remained statistically significant (adjusted RR, 0.89; 95% CI, 0.79–0.99; p = 0.03).

4. DISCUSSION

4.1. Principal findings

In this retrospective cohort analysis of Medicaid beneficiaries across six states, participation in a virtual maternity care program was associated with significant reductions in adverse neonatal outcomes that correlated with the duration of engagement. Our primary analysis indicates that sustained engagement (≥3 months) is associated with a 24.4% reduction in the adjusted risk of preterm birth, a 12.5% reduction in NICU admissions, and a 26.3% reduction in NICU length of stay. We observed a pattern of increasing benefit with longer program participation: while engagement of at least 1 month was associated with an 11.0% reduction in preterm birth, extending engagement to 3 months or more was associated with more than double the magnitude of effect. These findings suggest that virtual care models integrating clinical, behavioral, and social support may be associated with lower risk in high‐need Medicaid populations, particularly when engagement is sustained over time.

A central challenge in evaluating engagement duration is distinguishing the effect of the intervention from the natural accumulation of days in longer pregnancies. To address this, we employed two distinct analytical strategies. First, our primary outcome models utilized a doubly robust approach that explicitly adjusted for gestational age at enrollment, ensuring that comparisons statistically adjusted for when a patient entered the cohort. Second, to rigorously test the robustness of the association, we conducted a sensitivity analysis restricted to beneficiaries enrolled or rostered prior to 28 weeks’ gestation. In this sub‐cohort, all individuals, regardless of birth timing, possessed the theoretical temporal opportunity to achieve the maximal engagement definition (≥3 months, to 40 weeks’ gestation). In this restricted analysis, the protective association between high engagement and reduced preterm birth remained statistically significant. This supports the validity of the observed relationship between sustained engagement and improved outcomes.

4.2. Mechanisms

Multiple complementary mechanisms likely explain longer exposure associations. Many complications evolve gradually over weeks (chronic or gestational hypertension, gestational diabetes, perinatal mood and anxiety disorders). Continuous monitoring through remote physiologic data, proactive outreach, and 24/7 asynchronous access creates multiple detection opportunities that episodic visits may miss. Additionally, building trust for behavior change requires time and repeated interactions; patients engaged ≥3 months typically have 12+ contacts with the care team. Furthermore, longer engagement provides more opportunity to address multiple interrelated factors (medical conditions, mental health, social determinants) simultaneously. Lastly, immediate clinical care via text messaging (median 5‐min response time) with escalation to same‐day video visits creates responsiveness impossible in traditional models, where national audits show average wait times for new obstetric appointments now exceed 30 days [17].

4.3. Implementation implications

Achieving the observed results requires early enrollment and sustained engagement. To ensure 3 months of exposure before a term delivery, enrollment must occur by 26 weeks’ gestation. This underscores the need for proactive identification using all data sources available. Sufficient time must also be given to allow for scaling member enrollment.

Finally, the continuous, high‐touch engagement model demonstrated here faces structural barriers within traditional fee‐for‐service frameworks, which typically limit reimbursement to discrete billable encounters, and pay minimally for text‐based care. Such models may inadvertently disincentivize the longitudinal, asynchronous interactions required for effective remote management. Alternative value‐based arrangements, such as per‐member‐per‐month capitation or bundled pregnancy episode payments, may better align financial incentives with interventions such as this [18].

4.4. Comparison with literature

The 24.4% reduction in preterm birth observed in the high‐engagement cohort exceeds effects typically reported in broader telehealth literature. While previous systematic reviews have noted that virtual interventions improve patient satisfaction, evidence regarding their ability to reduce preterm birth has been mixed, often limited by heterogeneity in intervention design. Our findings align with the effect sizes seen in group prenatal care models (approximately 18% reduction in meta‐analysis) [19], suggesting that virtual care, when delivered with sufficient duration and clinical depth, may replicate some of the protective benefits of intensive in‐person support models, but with the added advantage of scalability to geographically dispersed populations.

4.5. Strengths and limitations

Strengths of this study include the large, geographically diverse sample and the use of rigorous methods to minimize confounding. We utilized doubly robust estimation, combining inverse probability weighting with regression adjustment. This was critical because, as shown in our balance tables, propensity weighting alone left residual imbalances in gestational age at enrollment and insurance start timing.

Limitations include the observational nature of the study, which precludes definitive causal assertions. While we adjusted for a wide array of clinical and social variables available in claims data, unmeasured confounders, such as patient motivation or health literacy, may persist. Additionally, preterm birth, NICU admission, and length of stay are inherently correlated outcomes. However, the consistent engagement duration relationship observed across all three measures reinforces the robustness of the association. Lastly, reliance on administrative claims data limits our ability to analyze granular clinical details, such as specific blood pressure readings or depression severity scores, which could further elucidate the mechanism of action. Furthermore, administrative claims data lack the clinical granularity to reliably differentiate between spontaneous preterm labor and medically indicated preterm birth (e.g., for preeclampsia). Finally, engagement was operationalized as duration (months) rather than intensity (frequency of interaction); future research should investigate how the frequency and type of interactions modify these outcomes.

5. CONCLUSION

This study provides evidence that a virtual maternity program is associated with reductions in preterm birth and NICU utilization among Medicaid beneficiaries. The findings were robust to sensitivity analyses accounting for the timing of enrollment, and the observation of greater benefit with longer engagement reinforces the potential value of sustained care. As payers and policymakers seek scalable solutions to maternal health challenges, these data suggest that virtual care models capable of maintaining patient connection over months, rather than weeks, may offer meaningful clinical value.

CONFLICT OF INTEREST STATEMENT

All authors with the exception of M.P. report employment at Pomelo Care and equity ownership in Pomelo Care.

ETHICS STATEMENT

This study was approved by an institutional review board with a waiver of informed consent and waiver of HIPAA authorization, as the research involved secondary analysis of existing administrative data, posed minimal risk to participants, and could not practicably be conducted without the waivers. All data were handled in accordance with HIPAA and HITRUST standards.

DATA AVAILABILITY STATEMENT

De‐identified aggregate summary data are available upon reasonable request to the corresponding author, subject to data use agreements with health plans and compliance with regulations including HIPAA.