Understanding and Addressing Severe Maternal Morbidity Among Black Women in Texas: Findings from a System Dynamics Group Model Building Study

Abstract

Introduction

To understand the dynamically complex causes of severe maternal mortality (SMM) among Black women in Texas, this study sought to answer the following questions: 1) What are the most important factors and forces that shape SMM these outcomes?; 2) How have those factors and forces changed over time, and what caused them to change?; 3) How are those factors and forces related to one another over time, perhaps in nonlinear or cyclical ways?; and 4) Where are the leverage points for introducing interventions?

Methods

Two participatory system dynamics group modeling building (SD GMB) sessions (2.5 total days) were held during March 2024 in Texas with 17 community-based stakeholders. Session activities elicited key variables, behavior-over-time graphs, causal loop diagrams, and targets for action.

Results

Stakeholders identified 351 key variables: 72 results, 83 roles, 74 relationships, 43 rules, and 79 resources. Cumulatively, 75 themes emerged across these variables. Stakeholders generated 16 behavior-over-time graphs that exhibited high degrees of uniqueness. The two causal loop diagrams featured 15 reinforcing and 27 balancing loops, and 31 reinforcing and 6 balancing loops, respectively. Finally, 36 targets for action were identified.

Conclusions

This study provided insights into the most important factors that shape SMM outcomes, how those factors may have changed over time, what may have caused them to change, and how they function in nonlinear and circular ways. Initial interventions were also identified. These findings bolster complex systems-grounded approaches, such as SD GMB, in maternal health by demonstrating their value for generating novel insights.

Introduction

Maternal mortality and morbidity continue to be significant public health issues in the United States (U.S.) [1]. Despite its number one ranking in per capita healthcare expenditures worldwide, along with its numerous advances in medical and public health science, maternal health outcomes in the U.S. fare poorly compared to other high-income countries [2, 3]. Especially troubling are increases in the U.S. maternal mortality rate in recent years, despite 80% of these cases being considered preventable [4, 5]. However, severe maternal morbidity (SMM), which refers to unexpected outcomes that occur during or shortly after labor and delivery that result in significant short-term or long-term consequences for a woman’s health, may be an especially important maternal health outcome [6]. Not only is SMM a key risk factor for maternal mortality, it also occurs much more frequently than maternal mortality, as there are 75 to 100 SMM cases for every maternal death [6, 7].

In the U.S., Black women experience excessively high maternal mortality rates that have persisted for over a decade and hold true even when controlling for other factors (e.g., socioeconomic status [SES]) [1, 8]. Similar outcomes exist regarding SMM, as Black women have the highest prevalence overall and for 22 of the 25 indicators used by the Centers for Disease Control and Prevention [9, 10]. As with maternal mortality, these outcomes have persisted for years and remain after controlling for factors such as SES [1, 11]. Numerous factors across levels of influence have been implicated as key drivers of the adverse maternal health outcomes – including SMM – that Black women experience. Proximal factors include clinical risk factors, such as advanced maternal age, birth history, pre-existing medical conditions, and health behaviors (e.g., smoking during pregnancy) [12, 13]. Healthcare factors, such as provider and facility characteristics, are also important [14]. However, many of the most important factors are believed to be rooted in historical structural forces that exert profound downstream impacts on many of the social determinants of health, such as by shaping macrostructural policies and systems [1, 15-20].

To investigate these multilevel factors and explicate their roles in birthing outcomes, maternal health research is predominantly grounded in linear and reductionist approaches [21]. Although these approaches have led to many insights into the causes of maternal health and have contributed to major clinical advancements [22, 23], the aforementioned outcomes experienced by Black women indicate that novel approaches may be necessary. While there is growing consensus that maternal health outcomes are caused by complex interactions of multilevel forces, linear and reductionist approaches cannot adequately study phenomena that are generated by nonlinear, dynamic, and emergent causal structures embodied in feedback loops that constitute such complex systems [24, 25]. These shortcomings in understanding the dynamic complexity of maternal health outcomes may also reduce the ability to meaningfully intervene in ways that lead to sustainable population-level impacts among Black women; for example, it is known that systems may respond to interventions through unexpected adaptations that lead to ineffective or even exacerbating impacts, which is a phenomenon termed “policy resistance” [24, 26].

In contrast to linear and reductionist approaches, complex systems approaches are adept at gaining insights into dynamically complex systems and delineate their causal feedback structures [27-31]. System dynamics modeling (SDM) represents a complex systems approach that has provided robust frameworks for understanding health outcomes [32-34]. This approach is perhaps most powerfully leveraged for studying feedback loops, which are articulated as processes whereby a change or response to a causal factor either reinforces the change (positive feedback loop) or stabilizes it (balancing feedback loop) [35, 36]. Although SDM has been underutilized in the context of maternal health, extant examples have often employed participatory approaches to capture the key underlying forces of the dynamically complex systems that shape these outcomes [33, 37, 38]. Such participatory approaches – referred to as system dynamics group model building (SD GMB) – are centered on collaboration with stakeholders within the language and techniques of SDM to develop a shared understanding of a problem (as an SD model) and identify potential solutions [39-41].

This paper describes the findings of a recent SD GMB study that sought to uncover the dynamically complex causes of SMM outcomes among Black women and identify interventions. The focus of this study was the state of Texas – the 2^nd most populous U.S. state [42] – where trends in maternal health among Black women parallel those that exist nationally, and indeed may be worse [14, 43, 44]. Texas has also experienced an increase in maternal mortality in recent years; however, 80% of these cases are considered preventable [14, 45]. These outcomes may be driven by unique contextual factors within the state; for example, reproductive-age women in Texas have the highest uninsured rate in the U.S., and the state ranks last nationally in access to preventive healthcare and maternal and reproductive care services [46, 47]. Nearly half of the counties in Texas are also characterized as maternity care deserts, and the state overall faring poorly in access to specific services (such as family planning) compared to the U.S. overall [48]. Black women in Texas have the highest rate of delivery hospitalization involving SMM overall, and SMM rates have nearly doubled in recent years [14, 45]. The study sought to answer the following questions: 1) What are the most important factors and forces that shape SMM among Black women in Texas?; 2) How have those factors and forces changed over time, and what caused them to change?; 3) How are those factors and forces related to one another over time, perhaps in nonlinear or cyclical ways?; and 4) Where are the leverage points for introducing interventions?

Methods

Two in-person SD GMB sessions (2.5 total days) were held during March 2024. Potential SD GMB stakeholders were identified using an iterative process that focused on: 1) determining which perspectives (e.g., birth workers) were most important to understanding the problem; 2) identifying potential stakeholders who could provide those perspectives; and 3) prioritizing the initial list of potential stakeholders for invitations. Twenty stakeholders accepted invitations to participate in the SD GMB activities, with 17 ultimately participating. These stakeholders represented a diverse array of perspectives on SMM outcomes among Black women in Texas. Their roles/titles, demographic characteristics, and sessions attended were ascertained through pre-session surveys and sign-in sheets and are presented in Table 1. All stakeholders completed informed consent forms. This study was approved by the Institutional Review Board at The University of Texas at Arlington (2024-0168).

Table 1.

Overview of System Dynamics Group Model Building Stakeholders

Stakeholder	Role/Title	Racial and Ethnic Identity	Gender Identity	Self-Reported Lived Experience	Session(s) Attended
1	Program Analyst	Non-Hispanic Black	Woman	No	1 and 2
2	Attorney	Non-Hispanic Black	Woman	No	1 and 2
3	Father, Community Expert, and Child Development Specialist	Non-Hispanic Black	Man	No	1 and 2
4	Research Scientist	Non-Hispanic Black	Woman	No	1 and 2
5	County Public Health Manager	Non-Hispanic Black	Woman	No	1 and 2
6	Labor and Delivery Nurse	Non-Hispanic Black	Woman	No	1
7	Father, Community Health Worker, and Community Expert	Non-Hispanic Black	Man	Yes	1 and 2
8	CEO	Non-Hispanic Black	Woman	Yes	1 and 2
9	Executive Director	Non-Hispanic Multi-Racial	Non-Binary	Yes	1
10	Research Analyst	Non-Hispanic Black	Woman	No	1 and 2
11	Community Expert	Non-Hispanic Black	Woman	Yes	1 and 2
12	Associate Professor	Non-Hispanic White	Woman	No	1 and 2
13	Environmental Commissioner	Non-Hispanic Black	Woman	No	1 and 2
14	Chair of OB/GYN Department	Non-Hispanic Black	Woman	No	1 and 2
15	Birth Justice Program Coordinator and Full Spectrum Doula	Non-Hispanic Black	Woman	Yes	1 and 2
16	CEO/Community Consultant	Non-Hispanic White	Woman	No	1 and 2
17	HIV Programs Director	Non-Hispanic Black	Woman	Yes	1 and 2

The two SD GMB sessions consisted of several group activities that followed well-documented procedures for maximizing effectiveness and research value (i.e., ‘scripts’) [49-55]. The scripts were adopted and adapted (e.g., by extending the amount of time on activities) based on the research questions and contextual factors, as well as from insights from a prior SD GMB pilot study conducted by the research team [21]. An overview of each major research activity is described in the sections below, with a compiled agenda that summarizes both of the SD GMB sessions provided in Table 2.

Table 2.

System Dynamics Group Model Building Agenda

Time	Activity
Session 1 (Day 1)
8:00am-8:45am	Introductions; overview of agenda, project activities and project purpose; development of community guidelines
8:45am-9:05am	Concerns and hopes
9:05am-9:15am	Biobreak
9:15am-10:50am	Variable elicitation (The 5 R’s)
10:50am-11:00am	Biobreak
11:00am-11:50am	Behavior-over-time graphs
11:50am-12:00pm	Model boundary chart
12:00pm-12:45pm	Lunch break
12:45pm-1:15pm	Presentation of conceptual causal loop diagram
1:15pm-1:35pm	Initiation and elaboration of causal loop diagram
1:35pm-1:45pm	Biobreak
1:45pm-3:30pm	Causal loop diagramming in small groups
3:30pm-4:00pm	End-of-day recap; stakeholder reflections; mindfulness exercise
Session 1 (Day 2)
8:00am-8:30am	Introductions; overview of agenda; reminder of project activities and project purpose; review of community guidelines
8:30am-8:45am	Review of previous day’s outputs and elicitation of additional input
8:45am-9:50am	Causal loop diagramming in small groups
9:50am-10:00am	Biobreak
10:00am-11:00am	Causal loop diagramming in small groups
11:00am-11:30am	Small group causal loop diagram presentations
11:30am-12:00pm	End-of-session recap; stakeholder reflections; mindfulness exercise
Session 2 (Day 1)
8:00am-8:15am	Introductions; overview of agenda; reminder of project activities and project purpose; review of community guidelines
8:15am-8:30am	Review of previous session’s outputs and elicitation of additional input
8:30am-9:50am	Causal loop diagramming in small groups
9:50am-10:00am	Biobreak
10:00am-12:00pm	Causal loop diagramming in small groups
12:00pm-12:45pm	Lunch break
12:45pm-1:20pm	Review and prioritization of small group causal loop diagrams
1:20pm-2:00pm	Targets for action
2:00pm-2:10pm	Biobreak
2:10pm-3:00pm	Wall of evidence for small group causal loop diagrams
3:00pm-3:30pm	Small group causal loop diagram presentations
3:30pm-4:00pm	End-of-session recap; stakeholder reflections; mindfulness exercise

5 R’s

Using the ‘5 R’s’ framework [56], stakeholders identified variables that they believed to be relevant to the problem across five domains: 1) results (what would success in addressing SMM among Black women in Texas look like?); 2) roles (who affects or can affect changes in the results, and who is affected or can be affected by changes in the results?); 3) relationships (what are the important relationships between the roles that affect, or affected by, the results?); 4) rules (which formal and informal rules affect the results or our ability to act to achieve them?); and 5) resources (what resources are available to improve the results?). Stakeholders were provided these definitions for each R, and they were then free to share any variable that they felt were important to share in the context of the problem through the process described below.

For each of the R’s, the process started with a divergent task, where stakeholders were asked to list as many variables as they could on separate sticky notes and then prioritize the variables they had identified. Then, as a convergent task, stakeholders took turns sharing their highest-priority variables in a round-robin fashion, and other stakeholders were asked to share if they had a similar variable as each one was shared. At the same time, the sticky notes that recorded each variable were collected, posted, and preliminarily themed into clusters in real-time. Preliminary themes were then discussed at the conclusion of the 5 R’s activity, and stakeholders were provided the opportunity to provide feedback on these themes to ensure they were accurate.

Model Boundary Chart

To prioritize the outputs of the 5 R’s for emphasis during subsequent SD GMB activities (e.g., causal loop diagramming), stakeholders were asked to vote using colored dots based on the following guiding question: “Which of the variables in each ‘R’ are the most important for understanding and/or addressing SMM among Black women in Texas?” For each of the 5 R’s, stakeholders were given one green (the most important item in each R), one yellow (the second-most important item in each R), and one red dot (the least important item in each R) to be placed on the corresponding sticky notes that they wished to vote for.

Behavior-Over-Time Graphs

To capture historical changes of the key variables of interest, all stakeholders created behavior-over-time graphs (BOTGs) of SMM rates among Black women in Texas (on the y-axis of the BOTG). Stakeholders then engaged in the divergent task of creating their own BOTG by following four steps: 1) determining the beginning time, which is how far back in time is needed to understand how the outcome has changed over time (the ‘time horizon’); 2) graphing how they believed the outcome has changed over time, since the beginning time; 3) graphing their beliefs on how they hoped, expected, and feared the outcome may change over the next 20 years; and 4) identifying the reasons they believed the outcome has changed or will change over time using notations. Then, based on the question: “Which BOTG best represents the dynamic characterization of SMM among Black women in Texas?”, stakeholders were given one blue dot each and asked to vote by placing the dot on the corresponding BOTG that they wished to vote for.

Causal Loop Diagrams

A causal loop diagram (CLD) is an informal causal map of a system that explicitly represent stakeholders’ mental models and often features key feedback loops [40]. As a SD modeling technique, a CLD consists of several ‘building blocks’, including: 1) variables; 2) links between variables, which indicate that, all else remaining equal, a change in the first variable (arrow tail) leads to a change in the second variable (arrow head); 3) polarities in the links, with a plus (+) sign indicating a change in the first variable causes a change in the same direction in the second variable, and a minus (-) sign indicating a change in the first variable creates a change in the opposite direction in the second variable; and 4) reinforcing (R) and balancing (B) feedback loops [35].

During the SD GMB process, causal loop diagramming activities took place across three steps. First, the rationale and function of CLDs was presented, which was accompanied by an example CLD developed by the research team that portrayed population dynamics (births and deaths). Second, the entire group of stakeholders worked together with the research team to develop a preliminary CLD ‘seed model’ in real-time, which was intended to demonstrate how mapping the mental models of stakeholders would take place during the subsequent activities and served as a starting point for developing full CLDs for the problem. Finally, the full group broke out into two ‘small groups’ (henceforth referred to as ‘small group 1’ and ‘small group 2’) to iteratively develop the focal CLDs for the problem, which was the most time- and effort-intensive portion of the SD GMB process and occurred across multiple days and sessions.

Targets for Action

To explore potential leverage points for introducing interventions to address the problem (and grounded by the CLDs that emerged), ‘targets for action’ were elicited by stakeholders within each small group. As with the 5 R’s task, the process started with a divergent task, where stakeholders were asked to list as many targets for action as they could identify on their small group’s CLD, each on a separate sticky note, and then prioritize the ones they had identified. Stakeholders were also asked to rate both the cost and the effectiveness of each action they identified, on a scale of 1 (low) to 10 (high), and to record these ratings on each sticky note. Then, as a convergent task, stakeholders took turns sharing their highest-priority targets for action in a round-robin fashion, and other stakeholders were asked to share if they had a similar target for action as each one was shared. At the same time, the sticky notes that recorded each target for action were collected, posted, and grouped onto x and y coordinates based on their cost (x-axis) and effectiveness (y-axis).

After identifying targets for action, stakeholders were then asked to vote on the final list, using green, yellow, and red dots, meaning “most likely to reduce SMM among NHBW [non-Hispanic Black women] in Texas”, “second most likely to reduce SMM among NHBW in Texas”, and “least likely to reduce SMM among NHBW in Texas”, respectively.

Data Analysis

5 R’s

Manifest content analysis (MCA) [57, 58] was employed to analyze the 5 R’s because it aligns well with one goal of this analysis: organizing stakeholders’ variables into thematic categories. MCA was also appropriate for this analysis because the responses were explicit, allowing the content’s meaning to be readily understood at the ‘surface’ level of the text [58]. MCA is a method of analyzing text-based data that focuses on explicit content, taking the meaning of words and phrases at face value [57]. It is an approach that involves counting and categorizing the explicit content based on what is literally present and staying as close to the text as possible [59]. A well-established analytic method for MCA was employed in the current study [57, 59].

First, it was determined that the unit of analysis for coding would be the words or phrases provided by stakeholders on the sticky notes, which were subsequently analyzed verbatim. All responses from the sticky notes were typed into an Excel file and organized within the corresponding R domain (results, roles, relationships, resources, or rules). Second, two coders independently read and re-read all responses in the Excel file to immerse themselves in the data. For an initial review, the coders reviewed 10 responses per R category (n = 50) and then held a meeting to ensure consistency. Next, after reviewing responses within each R domain, coders independently created a preliminary list of codes (also referred to as a coding scheme) and applied the preliminary codes to the responses. Each coder used a table template in Excel with three columns (code, response, and theme). In the third step, the two coders convened to review and discuss their preliminary list of codes and responses. The objective of this meeting was to reach consensus on how responses were coded, ensure the accuracy of the terminology used for codes, and discuss preliminary thematic categories. Although infrequent and typically limited to differences in wording/terminology (e.g., “policymakers” vs. “politicians”), the two coders managed disagreements when they did occur through consensus. Any disagreements for which consensus could not be reached were to be resolved by a third-party tiebreaker from the research team, but ultimately this was not necessary. In the final analytic step, the two coders created a table that included themes, a brief description of the theme, the number of responses within the theme, and illustrative responses for reporting purposes. In alignment with the principles of trustworthiness and to promote confirmability (or impartiality of the findings) [60], the coders used reflexive journaling [60], and to promote dependability (by assuring reliability via audit trails), the coders created a final comprehensive table in Excel that includes all themes, associated codes, and corresponding responses. For each of the R categories, the top three themes (based on the number of variables per theme) are reported in the Results section.

Model Boundary Chart

To analyze the model boundary chart, scores were assigned to stakeholders’ votes on each variable, with each green dot receiving two points, each yellow dot receiving one point, and each red dot receiving zero points. Cumulative scores were then calculated for each variable. The top three highest-scored variables in each domain (results, roles, relationships, resources, and rules) are provided in the Results section verbatim (as written by the stakeholder on the post-it note).

Behavior-Over-Time Graphs

The BOTGs that emerged from stakeholders had widely divergent time horizons and reasons for changes in trends over time (notations). As a result, thematic analysis using MCA was not practical, as few consistent themes emerged from the BOTGs. Instead, BOTGs were qualitatively evaluated based on the four attributes previously described (the beginning time; how the outcome has changed over time; how stakeholders hoped, expected, and feared the outcome may change over the next 20 years; and notations that describe trends). To demonstrate the insights elicited from this activity, the BOTG with the most stakeholder votes is described in the Results section, which includes verbatim annotations (as written by the stakeholder on the BOTG).

Causal Loop Diagrams

The initial causal loop diagrams that emerged from the two small groups were quite complex, consisting of dozens of variables and feedback loops. Therefore, the research team simplified these diagrams for reader coherence by: 1) removing several variables and connections that were exogenous (i.e., not embedded within feedback loops); 2) removing variables and connections that provided unnecessary detail that were endogenous (i.e., embedded within feedback loops) but not essential for describing causal mechanisms; and 3) combining endogenous variables that captured analogous causal pathways within feedback mechanisms. Additionally, to improve the readability of the CLDs, excessively long arrows were removed and replaced with duplicate variables (identified by carrots at the start and end of variable names in the diagrams) and shorter arrows. The research team also refined the CLDs by modifying variable names to ensure that they clearly and consistently expressed directionality in causal influence. These modifications were iteratively discussed among the research team and then member-checked with a subset of SD GMB stakeholders. Because these diagrams include an enormous amount of complexity, a full description of feedback loops is beyond the scope of this paper; instead, analysis focused on identifying broad clusters, or feedback mechanisms, that emerged during the activity and represent broader insights into key causal feedback structures, which are described in the Results section. To provide more in-depth depictions of these rich diagrams, one reinforcing loop and one balancing loop is described in the Results section for each of the two small group CLDs.

Targets for Action

As with the BOTGs, thematic analysis was not practical for targets for action due to the lack of consistent themes. Instead, analysis focused on two steps. First, cost and effectiveness scores for each target for action was transformed for analysis, with 0-5 categorized as “low” and 6-10 categorized as “high”. This resulted in each target for action falling within one of four categories: 1) high cost, high effectiveness; 2) high cost, low effectiveness; 3) low cost, high effectiveness; and 4) low cost, low effectiveness. Three of the targets for action did not include stakeholder ratings on cost and/or effectiveness, and these were therefore excluded from this analysis. Targets for action are reported in the Results section based on their distribution within these four categories.

Targets for action were also analyzed based on the number of votes they received, using the same scoring system as the model boundary chart (green = two; yellow = one; red = zero). Cumulative scores were then calculated for each target for action, and the top three highest-scored targets for action are provided verbatim in the Results section (as written by the stakeholder on the post-it note).

Results

5 R’s and Model Boundary Chart

Stakeholders identified a total of 351 variables across the 5 R’s: 72 results, 83 roles, 74 relationships, 43 rules, and 79 resources. Cumulatively, 75 themes emerged across these five categories: 11 themes for results, 13 themes for roles, 33 themes for relationships, 5 themes for rules, and 13 themes for resources.

The top 3 themes within the results R were as follows (with number of variables per theme included in parentheses): 1) Maternal Health Outcomes (15); 2) Education/Training (15); and 3) Healthcare Systems (11). Based on stakeholder votes, the specific variables receiving the highest scores were (verbatim): 1) Fewer overall Black maternal deaths); 2) trusting Black women; and 3) overall more pleasant birthing experiences.

Among roles, the top 3 themes were: 1) Healthcare Workers/Providers (21); 2) Policymakers (14); and 3) Black Families (13). Specific variables in this category receiving the highest scores were: 1) Black mothers; 2) several tied for second (e.g., mothers; policymakers; payers of insurance); and 3) two tied for third (Black OBGYN; education institutions).

With 33 themes, the highest number of themes emerged from the relationships R, with the top 3 being: 1) Black Women and Birthing People and Healthcare Workers/Providers (14); 2) Black Women and Birthing People and Black Families (11); and 3) Black Women and Birthing People and Policymakers (6). Specific variables receiving the highest scores were: 1) Black birthing people and policymakers; 2) constituents and policymakers; and 3) several tied for third (e.g., Health providers and Black reproductive age people; OBGYN and L&D nurses; doula and hospital birth team).

Rules had fewest themes (5), with the top 3 being: 1) Social/Cultural Norms (18); 2) Healthcare Norms (11); and 3) Healthcare/Hospital Policies (7). The variables within this category that received the highest scores were: 1) Discounting Black voices; 2) Hospital policies affecting provider practice (where and how); and 3) Hospital policies (e.g., massive blood transfusion; transfer to higher level care).

Finally, resources featured the following top 3 themes: 1) Community (24); 2) Income (13); and 3) Healthcare Access/Quality (9). The top three highest-scoring variables in this category were: 1) Parenting in community; “It takes a village”; 2) guaranteed minimal income; and 3) several tied for third (e.g., culturally competent providers; affordable, safe, and stable housing; community-based education and parental training).

Behavior-Over-Time Graphs

The 16 BOTGs of SMM rates among Black women in Texas exhibited a high degree of diversity across the four attributes analyzed. First, regarding time horizons, many started as early as 1490s, with many indicating “Pre-1619” or “1619”, but one was as recent as 2016. In terms of historical trends, BOTGs demonstrated substantial variation, with most showing an initial increase, but some showed decreases in more recent years while others showed increase in more recent years; further, some BOTGs show ebbs and flows. For future trends, the ‘feared’ trend lines all showed increase, and ‘hoped’ trend lines showed decrease; however, ‘expected’ trend lines exhibited had a high degree of variation. Finally, explanations for changes in historical and/or future trends (notations) spanned a wide range: Many BOTGs included notations that referred to social forces/policies, including slavery, Jim Crow laws, New Deal laws, and the War on Poverty. Several other BOTGs included notations related to healthcare, such as birthing conditions, health care practices, and hygiene. Other BOTGs included more unique notations (e.g., pregnancy rates; the involvement of fathers; communities).

The specific BOTG that received the most votes is provided in Supplement 1 and featured a time horizon that started in 1619. Historically, this BOTG demonstrated a gradual decrease in SMM rates among Black women in Texas until roughly the 1800’s, after which there was a gradual increase until present day. For future trends, the ‘feared’ line was sharply up, the ‘hoped’ line was sharply down, and the ‘expected’ line was slightly up. This BOTG included several notations to explain historical trends, including slavery, Black self-sufficiency, integration, and systemic racism. Notations were also included for future trends: ‘feared’ was explained by “systematic racism will not be eradicated but become more engrained”, ‘hoped’ was explained by “Black voices are heard” and “permanent and effective changes are made/implemented”, and ‘expected’ was explained by “history will repeat with short/small improvements = temporary fixes”.

Causal Loop Diagrams

The causal loop diagrams that emerged from the two small group processes are provided in Figure 1 and Figure 2. The outcome of interest (SMM rates among Black women in Texas) is identified in bold font in each figure. The CLD that was produced by Small Group 1 (Figure 1) included 15 reinforcing and 27 balancing loops, and nine clusters (i.e., feedback mechanisms) are embedded within this CLD: 1) proximal consequences of SMM for Black women and their families; 2) mental health/stress; 3) racism and discrimination; 4) healthcare bias/access/quality/utilization; 5) neighborhood characteristics; 6) criminal justice system; 7) education; 8) wealth; and 9) community advocacy.

Fig. 1: Causal Loop Diagram from Small Group 1.

Fig. 2: Causal Loop Diagram from Small Group 2.

Within the neighborhood characteristics and wealth clusters, a reinforcing loop (R9) depicts how investment in neighborhoods can create generational wealth: An increase in housing ownership leads to higher quality of neighborhoods, which then increases the value of homes and increases generational wealth, which further increases housing ownership. However, balancing loop B17 shows how this process has tradeoffs, as those increasing levels of housing ownership and subsequent higher value of homes can, in turn, reduce housing ownership (e.g., by making home ownership unaffordable). The components of these loops within the CLD are bolded in Figure 1.

The CLD that emerged from Small Group 2 (Figure 2) featured 31 reinforcing and six balancing loops, with eight feedback mechanisms within this diagram: 1) proximal consequences of SMM for Black women and their families; 2) mental health/stress; 3) historical legacy of discriminatory policies; 4) influence of multi-level racism on healthcare access/quality/utilization; 5) criminal justice system; 6) neighborhood characteristics; 7) media; and 8) community advocacy.

Within the proximal causes of SMM for Black women and their families, influence of multi-level racism on healthcare access/quality/utilization, and community clusters, a reinforcing loop (R8) shows how the consequences of SMM can be perpetuating: As SMM rates among Black women in Texas increase, long-term morbidity also increases, leading to decreases in household income, decreases in access to healthcare, and decreases in prenatal care entry/utilization, which then further increases SMM rates among Black women in Texas. However, balancing loop B4 shows how this vicious cycle does not escalate indefinitely: As SMM rates among Black women in Texas increase, community awareness also increases, and with that comes more community advocating and organizing, which leads to more funding/community partnerships and fewer discriminatory policies, resulting in improved prenatal care entry/utilization and reduced SMM rates among Black women in Texas. The components of these loops within the CLD are bolded in Figure 2.

Targets for Action

Cumulatively, 36 overall targets for action were elicited from stakeholders, with 25 emerging from small group 1 and 11 emerging from small group 2. Based on cost and effectiveness, these targets for action were grouped as follows: 1) 24 were high cost-high effectiveness; 2) 1 was high cost-low effectiveness; 3) 8 were low cost-high effectiveness; and 4) none were low cost-low effectiveness. Among the specific targets for action, the top three receiving the highest scores (based on stakeholder votes) were: 1) More support groups without child protective services (low cost-high effectiveness); 2) three tied for second (extended parental leave with pay [high cost-high effectiveness]; lower housing costs [high cost-high effectiveness]; extend maternity leave with pay [low cost-high effectiveness]); and 3) normalize additional care services in routine practice (high cost-high effectiveness).

Discussion

As SMM continues to burden Black women in Texas, novel insights into the multilevel factors that constitute the systems that shape these outcomes continue to be necessary to guide intervention efforts. The SD GMB approach employed in the current study provided such insights into the most important factors and forces that shape these outcomes, as well as how those factors and forces may have changed over time, what may have caused them to change, and how they may function in nonlinear and circular ways (as feedback loops). Based on this understanding, initial leverage points for interventions were also identified.

Identifying the Most Important Factors and Forces that Shape SMM among Black Women in Texas

A rich and diverse array of relevant variables emerged from the 5 R’s activity that shed light into the factors and forces that shape SMM among Black women in Texas. These variables were then prioritized during the model boundary chart activity to provide additional insights into their relative importance to the problem, and these variables were then emphasized by stakeholders during the development of the small group CLDs. This process resulted in a large and diverse set of variables that, together, indicate the breadth of factors relevant to the problem.

Regarding specific variables that were elicited, many of the most important variables were centered on healthcare, which was the richest type of variable across all five R’s and corresponds with the existing literature [14, 61]. The healthcare-related variables that were identified by stakeholders spanned multiple levels of influence, from individual birthing experiences to healthcare systems. Other variables have been established in the literature as well to varying degrees, such as those related to income and housing [62, 63]. However, other healthcare variables, such as payers of insurance (a role) and relationships between doulas and hospital birth team (a relationship), may be less prevalent in existing studies and thus warrant further investigation.

Demonstrating and Explaining Dynamic Behavior of SMM among Black Women in Texas

While the 5 R’s and model boundary chart activities provided rich insights into the causal factors and forces relevant to SMM among Black women in Texas, the BOTGs constituted insights into the dynamic of the problem. Although relatively diverse, these BOTG overall spoke to the importance of historical factors and forces that are distal to present-day maternal health outcomes among Black women in Texas and therefore to the need to widen the spatiotemporal ‘lens’ through which the problem is viewed and studied. For example, the time horizons of the BOTGs typically extended centuries back, many to roughly the early 17^th century. Further, the annotations included in these BOTG implicated historical policies and forces – often starting with slavery, and extending through other macrostructural policies and changes – as causal explanation of these trends. Through examination of the BOTGs, it appeared that establishing the temporal scope of the problem (i.e., the time horizon) also shaped how causality was inferred, with those BOTG with longer time horizons often focusing on more distal causal explanations for historical (and future) trends. However, even those BOTG with relatively more proximal time horizons (e.g., starting in the year 2000) included annotations that were more distal in influence (e.g., healthcare, housing, & food assistance programs). As new policies and other distal forces that may shape SMM rates among Black women dynamically unfold over time, approaches such as generating BOTG may be especially and uniquely valuable for ascertaining their effects over time.

Circular Causality in the Causes and Consequences of SMM among Black Women in Texas

The CLDs developed by the two small groups represented novel maps of the systems that shape SMM among Black women in Texas. These CLDs represent a cumulative output from the preceding SD GMB activities. For example, the 5 R’s and model boundary chart activities helped guide stakeholders into the systems thinking mindset needed to meaningfully contribute to building a CLD. By reflecting and prioritizing key variables within the roles, relationships, rules, resources, and results domains, stakeholders begin to surface key dynamics, patterns, and structures within the system, effectively preparing them to co-create a meaningful CLD. Similarly, creating the BOTG represented another step towards systems thinking by emphasizing the dynamic and spatiotemporally distal nature of the causes and outcomes related to the problem. The outputs of these activities also helped the SD GMB facilitation team develop targeted prompts that were later used as seeds to re-engage stakeholders during the small group CLD activity by serving as effective entry points to stimulate discussion and further progress with the system mapping process.

Although these two diagrams were developed independently, they shared several feedback mechanisms that may be interpreted as especially important to understanding the problem. The consistency of these mechanisms across both small groups may provide clues into latent ‘dominant’ feedback loops that drive the overall dynamics of the system [64]. Several of the variables within these overlapping mechanisms (e.g., proximal causes of SMM for Black women and their families; mental health/stress; neighborhood characteristics) are supported by the existing literature, such as lower incomes [65, 66], increased stress and weathering [67, 68], and poorer neighborhood characteristics [65, 66, 69, 70] experienced by many Black women.

However, there were key distinctions between the two small groups in how their CLDs emerged. For example, both CLDs include variables related to racism, which was expected and in-line with much of the literature [1, 17, 18]. However, small group 1’s CLD more generally depicted this factor as just ‘racism’, while small group 2’s CLD depicted racism as a distinct phenomenon at multiple levels (from systemic and structural & institutional, to interpersonal and internalized). These differences impacted how the CLDs themselves were developed, as small group 2’s CLD provided a more nuanced understanding of how each manifestation of racism influenced various elements of the system. At the same time, both CLDs fundamentally depicted the same mechanisms associated with racism that also corresponded with the existing literature, including healthcare [71], housing [72], and criminal justice [73]. Overall, compared to the literature, the CLDs represent a novel and holistic conceptualization of these factors and how they are interrelated, and thus these findings are wholly unique contributions to the maternal health knowledge base.

Leverage Points Within the Complex Systems that Shape and Perpetuate SMM among Black Women in Texas

The complex systems thinking that stakeholders had engaged in throughout the earlier activities led to improved recognition of the most important leverage points for introducing interventions, as well as the relative effectiveness of those interventions. These insights are critical, as policy resistance has been observed in existing prevention programming aimed at reducing undesirable maternal health outcomes among Black women [74]. In contrast, because targeting dominant feedback loops may be vital to interventions [75-78], the targets for action identified in the SD GMB approach employed in this study may represent high-leverage intervention points [34, 35]. Indeed, many of the targets for action were associated with variables within the feedback loops, such as lower housing costs, more community supports, and longer paid maternity leave. Further, the growth in systems thinking capacity among stakeholders during the course of the SD GMB activities resulted in initial intervention ideas that target key feedback loops through both structural changes (high cost-high effectiveness) and ‘low-hanging fruit’ (low cost-high effectiveness).

Future Directions for Research

Overall, it is hoped that the findings reported here may bolster further complex systems-grounded approaches in maternal health by demonstrating the value of SD GMB approaches for both identifying novel insights into longstanding maternal health problems and for integrating existing insights into holistic frameworks (e.g., causal loop diagramming). This is also in-line with recent calls within the maternal health literature for more frequent use of SD modeling approaches and for greater involvement of community stakeholders in such initiatives [79, 80]. Other complex systems-grounded approaches, such as agent-based modeling, are also encouraged, as they are amenable to participatory approaches [81] and have become more common within maternal health contexts [e.g., 82].

Although SD GMB approaches can serve as stand-alone qualitative studies [e.g., 21], they are often leveraged in especially powerful ways towards the development of quantitative SD simulation models [e.g., 83]. Accordingly, this study represents the first step in a broader research project currently underway that consists of multiple phases, with qualitative SD GMB outputs providing the foundation for a quantitative SD simulation model that is currently under development. The quantitative SD simulation model will then be used in conjunction with additional stakeholder input within a future series of structured activities for the identification of effective interventions.

Limitations

The study reported in this paper had four primary limitations. First, although many of the key findings correspond with related maternal health studies, some of the findings described in this paper may not be generalizable to other problem, population, or geographic contexts. Second, even with the specificity of the problem context (SMM rates among Black women in Texas), the investigation undertaken through SD GMB methods reported here was still challenging in terms of both combinatorial (the number of relevant factors and forces) and dynamic (the interactions among these factors and forces over time) complexity. Even with a significant amount of time allocated to SD GMB activities, as well as follow-up member checking, it is likely that additional insights would have been elicited with more time. Third, the findings reported here, and especially the CLDs, represent the verbatim output of the SD GMB activities. This does not mean that these findings are necessarily incorrect, but rather that they warrant caution in interpretation and likely need further refinement (e.g., through integration and/or simplification [40]) prior to translation into quantitative SD modeling. Finally, as this study is qualitative in nature, it is subject to many of the same limitations common to other qualitative designs, such as having a small sample size, although robust in the context of SD GMB conventional group sizes [84]; possibly lacking generalizability, as described in the first limitation; and requiring caution in terms of causal inference.

Conclusion

The SD GMB approach employed in this study shed light on the dynamically complex causes of SMM outcomes among Black women and initial interventions. Specifically, this approach led to understanding of the most important factors and forces that shape these outcomes, how those factors and forces may have changed over time, what may have caused them to change, and how they may function as interrelated feedback structures. In the context of these complex systems-grounded insights, initial leverage points for introducing interventions were also identified. It is anticipated that these rich insights will demonstrate the value of complex systems, SD modeling, and especially SD GMB in maternal health, as these participatory approaches can be leveraged in especially powerful ways towards the development of quantitative SD simulation models and the identification of high-leverage interventions.