Substance Use During Pregnancy and its Association with Maternal Health Among Women of Reproductive Age in Alabama: An Analysis Using the Three Delays Model and PRAMS Phase 8 Data

1. Introduction

Substance use during pregnancy presents substantial risks to maternal and neonatal health globally, contributing to adverse birth outcomes, including stillbirth, preterm birth, birth defects, and low birth weight [1]. Inadequate infrastructure challenges have made it difficult to find lasting solutions to a burgeoning public health problem that continues to threaten the social fabric of society, community safety, and the health and well-being of families. Because the burden of disease and disability associated with substance use during pregnancy evolves over a period, understanding and updating our knowledge and strategies on the structural and social drivers of this public health threat must be prioritized to improve outcomes for the population affected.

Even though significant efforts have been made over the past decades to lessen the risks associated with substance use during pregnancy among women of reproductive age, there are still substantial gaps in contextualizing the psychosocial factors that contribute to this problem and how to galvanize community action to address this ongoing health issue [2]. Interestingly, past research has shown that psychosocial and behavioral interventions, like contingency management, motivational support, and cognitive behavioral therapies, are crucial to addressing substance use during pregnancy [3]. Notwithstanding, designing and implementing innovative and sustained strategies that identify at-risk pregnant women early and get them to treatment remains a challenge, underscoring the need for information seeking during the prenatal period.

Geographic factors and cultural nuances can also influence the burden and disability associated with substance use during pregnancy across diverse care settings. The prevailing thought is that culture plays a crucial role in how we behave and communicate, and these intricate nuances affect our beliefs and views on substance and alcohol use [4]. This insight further supports a need to uncover the multidimensions of the interplay and interconnectedness of factors that influence pregnant women afflicted with substance use to seek care, reach care, and receive care. The knowledge gained can guide the development of robust strategies to improve outcomes for those most affected by this public health problem. Notably, opioid use among pregnant women is still deemed a significant public health concern [2], deserving priority attention and policy realignment.

To understand the gravity of this problem, data from 2010 to 2017 shows a 131% increase in opioid-related diagnoses at delivery hospitalization [2,5]. More crucially, opioid use disorder during pregnancy is known to be associated with serious adverse health outcomes, including preterm birth, stillbirth, maternal mortality, and neonatal abstinence syndrome [2]. These adverse health outcomes highlight the urgency and timeliness of this issue, demanding targeted research, implementation of sustained evidence-based practices, program evaluation, and monitoring to ensure improved health gains [4,6]. Given these circumstances, the Three Delays Model was employed in this research to evaluate and understand the complex relationships between substance use and maternal health outcomes and their mediating and moderating effects. This framework has the potential to offer new insights into the latent and observable factors concerning substance use during pregnancy and maternal health outcomes, thereby providing a fuller picture and the context-specific needs to prioritize. Based on the Three Delays Model, the research hypotheses are as follows:

• Higher levels of prenatal care barriers, particularly system-level and informational barriers, would be associated with poorer maternal health outcomes.
• Informational barriers would mediate the relationship between substance use and maternal health.

2. Methods and Statistical Analysis

Descriptive statistics were generated to summarize sample characteristics and variable distributions. Exploratory techniques, including multidimensional scaling (MDS) and hierarchical clustering, were performed to ascertain the grouping of prenatal care barrier items. Internal consistency was evaluated using Cronbach’s alpha and McDonald’s omega. Given the low reliability estimates for the composite measures [omega totals of 0.11 for Delay Seeking, 0.07 for Delay Reaching, and 0.14 for Delay Receiving], an IRT model was employed to derive the latent trait for informational barriers (theta_info). The inferential analyses included logistic regression to examine associations among substance use, prenatal care barrier constructs, and maternal health outcomes while controlling for demographic covariates. The logistic regression model was specified as:

logit[P(Y = poor)] = β₀ + β₁(smoker_status) + β₂(structural_barriers_new) + β₃(system_barriers_new) + β₄(theta_info) + β₅(Age_Group) + β₆(Race/Ethnicity) + β₇(Qn79_IncomeCategory) + β₈(BMI_Category).

Mediation analyses, utilizing bootstrapping (1,000 simulations), were employed to evaluate whether informational barriers mediated the relationship between substance use and maternal health. A structural equation model (SEM) was estimated using the WLSMV estimator in lavaan, with maternal health status treated as an ordered categorical variable. Multi-level covariates were converted to numeric form for model inclusion, and model fit was evaluated using indices such as the Comparative Fit Index (CFI), Tucker–Lewis Index (TLI), and Root Mean Square Error of Approximation (RMSEA). All analyses were performed in R version 4.4 using packages including dplyr, tidyr, psych, mediation, and lavaan.

3. Results

Table 2 below shows the frequencies and percentages for the key study variables. For example, 53.8% of respondents were classified as having good maternal health, while 46.2% had poor maternal health. In addition, 60.2% were categorized as non-smokers, and 39.8% were smokers. The table also displays the distributions for individual Qn21 items, income categories, BMI, age groups, and race/ethnicity.

To visually assess the underlying grouping structure for each of the items, a multidimensional scaling (MDS) plot was employed. As shown in Figure 1, the items clustered in a two-dimensional space, revealing several groupings that may correspond to distinct barriers. Items associated with financial or scheduling constraints, like Qn21b_Money and Qn21a_Appoint, appeared in one region, while items reflecting knowledge or secrecy barriers, like Qn21i_Unaware and Qn21j_Secret, appeared in another. These patterns provide initial evidence that these items may capture multiple underlying dimensions of prenatal care barriers, aligning partially with the Three Delays Model.

To determine whether different types of delays (seeking, reaching, and receiving care) have a significant relationship with maternal health status, bivariate analyses using chi-square tests were employed. The bivariate analyses revealed that the Delay Seeking factor was significantly associated with maternal health status (χ²(3) = 23.11, p < .001), while the Delay Reaching (χ²(1) = 2.59, p = 0.462) and Delay Receiving (χ²(1) = 4.42, p = 0.229) factors were not. These results show that, at the bivariate level, only variations in Delay Seeking are significantly related to differences in maternal health outcomes.

Expanding on these initial findings, a logistic regression model was employed to assess associations among substance use, prenatal care barrier constructs, and maternal health outcomes while controlling for demographic covariates. As shown in Table 3, system-level barriers (OR = 1.77, p = .046) and informational barriers, as measured by the IRT-derived latent trait (theta_info; OR = 2.42, p < .001), were significantly associated with adverse maternal health outcomes. However, structural barriers and substance use did not show significant direct effects.

These results suggest that, among the prenatal care barriers examined, system-level barriers or obstacles and informational barriers, as measured by the IRT-derived latent trait, are the most robust predictors of maternal health outcomes. Specifically, higher levels of system barriers increase the odds of poor maternal health by 77%, and a one-unit increase in informational barriers is associated with a 142% increase in the odds of poor health. Although substance use showed a trend toward significance, its association was not statistically robust after adjusting for other factors. Together, these findings underscore the importance of addressing systemic and informational barriers in interventions to improve maternal health outcomes.

A hierarchical clustering analysis was performed to identify distinct groups of prenatal care barriers. This method shows how specific barriers cluster together, offering insights into their relationships and guiding targeted perinatal interventions. Figure 2 depicts the hierarchical clustering dendrogram for the Qn21 items, which were hypothesized as representing different domains of prenatal care barriers. Several notable clusters emerged.

For example, Qn21b_Money and Qn21f_WorkSchool are grouped, showing a potential overlap in financial and work-related constraints. Likewise, Qn21g_MedicaidCard, Qn21e_Busy, and Qn21i_Unaware formed another cluster, revealing a set of logistical or knowledge-related barriers or issues. Conversely, Qn21j_Secret and Qn21k_NoPNC appeared as a distinct pair, potentially capturing personal or privacy concerns associated with pregnant women not seeking prenatal care.

These groupings largely align with the patterns observed in the Multidimensional Scaling (MDS) analysis shown in Figure 1. However, the dendrogram does not wholly separate the items into three discrete clusters corresponding to the Three Delays Model. Instead, the results suggest that some barriers, like money and work constraints, may cut across multiple theoretical domains. Composite scores for each cluster were created to understand their relationship to maternal health outcomes. Further steps involve refining these item groupings and examining how they relate to maternal health outcomes.

Following the hierarchical clustering analysis, the complex relationships between prenatal care barriers, substance use, and maternal health outcomes were further evaluated through mediation analysis. This approach allowed for an assessment of the indirect effects of informational barriers on the observed associations. The mediation analyses using 1,000 bootstrap simulations were performed to test whether informational barriers mediated the relationship between substance use and maternal health. Table 4 summarizes these findings.

The average causal mediation effect (ACME) was statistically significant (Estimate = 0.03, 95% CI [0.01, 0.05], p = .006), indicating that informational barriers significantly mediate the relationship between substance use and maternal health. The average direct effect (ADE) was not statistically significant (Estimate = –0.08, 95% CI [–0.16, 0.00], p = .056), and the total effect was non-significant (p = .232). The proportion mediated was –0.44. The negative value suggests that the mediated (indirect) effect operates in the opposite direction of the direct effect, a pattern that may indicate a suppression effect. In essence, while substance use tends to reduce the odds of poor maternal health (although not significantly so), the pathway through informational barriers contributes an effect in the opposite direction. These findings highlight the complex interplay of factors influencing maternal health outcomes and strengthen the importance of informational barriers in this relationship.

To expand on and confirm the findings from the logistic regression and mediation analyses, a structural equation model (SEM) was estimated. This thorough approach allows for the simultaneous examination of multiple relationships among the variables, providing a robust framework to understand the complex dynamics influencing maternal health outcomes. Table 5 reports the parameter estimates from a structural equation model testing whether informational barriers (measured by the IRT-derived latent trait, θ) mediate the relationship between substance use (smoker_status) and maternal health status (good vs. poor), controlling for structural and system barriers and demographic covariates.

The mediator equation revealed that smoker_status significantly predicted θ (a = 0.143, SE = 0.058, 95% CI [0.029, 0.257], p = .014). In the outcome equation, θ significantly predicted maternal health status (d = 0.530, SE = 0.079, 95% CI [0.375, 0.685], p < .001). However, the direct effect of smoker_status on maternal health (c′ = –0.199, SE = 0.111, 95% CI [–0.417, 0.019], p = .072) did not reach conventional significance.

The estimated indirect effect (a*d = 0.076, SE = 0.032, 95% CI [0.013, 0.139], p = .018) was statistically significant, indicating that informational barriers partially mediate the association between substance use and maternal health. The total effect was also non-significant (–0.123, p = .269). These results suggest that while the direct effect of substance use on maternal health is weak, a significant portion of the effect is transmitted indirectly through informational barriers.

To illustrate the discriminative properties of the informational barriers’ items, Figure 3 shows the item characteristic curves. For example, the item characteristic curves (ICCs) for Qn21i_Unaware, Qn21j_Secret, and Qn21k_NoPNC under a two-parameter logistic (2PL) IRT model. Qn21i_Unaware demonstrates a relatively flat slope, reflecting low discrimination. In contrast, Qn21j_Secret shows a sharply decreasing curve, consistent with its high negative loading and strong discrimination. Qn21k_NoPNC also depicts a steep slope, revealing moderate-to-high discrimination. These patterns suggest that Qn21j_Secret and Qn21k_NoPNC are particularly effective at distinguishing respondents across the latent trait of informational barriers, whereas Qn21i_Unaware contributes less to differentiating individuals. Despite the limited discrimination of Qn21i_Unaware, the items collectively capture a meaningful latent construct of informational barriers to prenatal care. These curves visually demonstrate how each item differentiates between various levels of the latent trait, further emphasizing the significance of informational barriers in understanding maternal health outcomes.

4. Discussion

This cross-sectional study is among the few that have examined the association between substance use during pregnancy and maternal health outcomes in Alabama using the Three Delays Model. While this framework has been extensively employed in maternal health research [7,8,9,10,16], its application to substance use during pregnancy remains limited. This study contributes to the literature by examining the association between substance use and maternal health outcomes through the lens of this framework, emphasizing the mediating role of informational barriers.

Barriers to prenatal care were conceptualized and categorized into three domains: structural barriers, informational barriers, and system barriers. The analysis showed that system-level barriers and informational barriers were significantly associated with adverse or poor maternal health outcomes [11,15,17,18,19,20]. In contrast, structural barriers and substance use (smoker_status) did not exhibit significant direct effects, as observed in this study.

Additionally, mediation analyses demonstrated that informational barriers, which were operationalized as an IRT-derived latent trait (theta_info), significantly mediated the relationship between substance use and maternal health outcomes, with approximately 44 percent of the effect transmitted indirectly. Likewise, the structural equation model (SEM) corroborated these findings, showing excellent model fit and robust parameter estimates. The methodological robustness of combining IRT and SEM is supported or validated by past research [21,22].

4.1. Implications for Theory and Practice

The study findings highlight the utility of the Three Delays Model in understanding prenatal care barriers and how they impact maternal and perinatal outcomes. For instance, the significant effects of system-level and informational barriers suggest that pregnant women experiencing substance use disorders face challenges related to navigating the healthcare system. Deficits in awareness about and access to prenatal health services and other community resources and informational needs during the perinatal period critically continue to impact maternal health outcomes.

Research has shown that low maternal health literacy significantly affects maternal and child health outcomes, including access to essential care and health services [23]. Likewise, language barriers and health communication anxiety among linguistic minorities exacerbate healthcare disparities and limit access to appropriate services [24]. Cultural norms and health-seeking behaviors also shape the utilization of healthcare services, particularly in diverse or rural settings [25]. Developing culturally tailored health education materials and initiatives has been shown to reduce disparities and improve outcomes in vulnerable populations [26]. Alabama could benefit from such efforts by investing in telehealth and mobile clinic initiatives. The mediating role of informational barriers implies that, even when substance use may not directly compromise maternal health, the lack of information or awareness about prenatal care can exacerbate adverse outcomes.

These results extend previous research by underscoring the complex interplay of behavioral and systemic factors.

In practical terms, the findings suggest that interventions aimed at improving maternal health in Alabama should prioritize enhancing the accessibility and quality of information about prenatal care and other community support services for those in need during the prenatal and postpartum periods. Referral and case management services should complement other public health evidence-based practices to sustain health gains. Addressing systemic issues like appointment scheduling and service availability should be prioritized in healthcare planning and policy development.

4.2. Study Limitations

Despite its strengths, this study has a few limitations that merit discussion. First, its cross-sectional design limits causal interpretations, as the associations identified do not necessarily imply causation. Second, the reliance on self-reported data from the PRAMS Phase 8 data set introduces the possibility of response bias, as participants may underreport or overreport behaviors and experiences. While this limitation is inherent to survey-based research, it is essential to interpret the findings with caution.

Third, the study findings are specific to Alabama and cannot be generalized to other states or regions. The design of PRAMS, including its implementation of state-specific core questions and sampling methodologies, inherently limits comparability across states. This methodological nuance highlights the importance of situating the findings within the context of Alabama’s unique demographic, socioeconomic, and healthcare characteristics. Lastly, the low reliability of some composite measures necessitated the use of advanced statistical techniques, such as item response theory (IRT), to derive latent traits. Although these methods address reliability concerns, they may obscure nuances within individual items. Future research should consider qualitative approaches or longitudinal designs to provide a deeper understanding of these prenatal care barriers and their influence on maternal health outcomes.

5. Conclusions

Altogether, this study demonstrates that system-level and informational barriers significantly influence maternal health outcomes among women who experience substance use during pregnancy. Addressing these barriers through policy changes, tailored interventions, and enhanced access to resources could improve prenatal care utilization and, consequently, lead to better maternal and infant health outcomes.